- 
+ 
@@ -54,7 +54,7 @@ It also does three things `curl` will not:
on every later request. `--continue` handles that; without tools it
strips the same field, because replaying it there just burns tokens.
- **Prices every call.** DeepSeek's disk KV-cache makes a cached input
- token **~30× cheaper** than an uncached one, and since 2026-08-16 the
+ token **30–50× cheaper** than an uncached one, and since 2026-08-16 the
hour of the day doubles the bill on top. Both splits are invisible
unless something reads `prompt_cache_hit_tokens`, knows the schedule
and does the arithmetic. This does, on every call, and keeps a local
@@ -171,7 +171,7 @@ is the key, the balance, the network, a proxy, or one endpoint:
$ deepseek check
https://api.deepseek.com
-ok GET /models 141ms deepseek-v4-flash, deepseek-v4-pro
+ok GET /models 141ms deepseek-flash, deepseek-v4-pro
ok GET /user/balance 126ms 18.48 CNY
ok POST /chat/completions 651ms 5 in / 1 out
ok POST /anthropic/v1/messages 590ms 5 in / 1 out
@@ -376,8 +376,8 @@ context cache saved ~$0.23 (1.7M of 2.2M prompt tokens replayed)
costs are estimates from the published USD rate card, not billed amounts
```
-That last line is the point. Cached input costs $0.007/M against
-$0.22/M for a miss in an off-peak hour — structuring prompts so the
+That last line is the point. On flash, cached input costs $0.003/M
+against $0.15/M for a miss in an off-peak hour — structuring prompts so the
stable part comes first is worth real money, and this is how you see
whether it worked.
@@ -470,7 +470,7 @@ rather than hides:
the same number at every level — and at low effort it is not there at
all, while the model still reasons:
- | `--effort` | flash | pro |
+ | `--effort` | V4 flash | pro |
| --- | --- | --- |
| `none` | +0 (thinking off) | +0 (thinking off) |
| `minimal`, `low` | **+0** | **+0** |
@@ -478,12 +478,15 @@ rather than hides:
| `max` | +92 | +79 |
Measured against the live API on 2026-08-05 at two prompt lengths,
- twice each. `deepseek tokens -e low` will show you the same thing for
+ twice each, on V4 Flash. That model was retired on 2026-09-10 and
+ `deepseek-flash` has not been re-measured: one probe on 2026-09-18
+ read +26 at `high`. `deepseek tokens -e low` will show you the same thing for
your own text. Two of those levels — `none` and `minimal` — appear in
no DeepSeek documentation at all; `none` disables thinking exactly as
`--think off` does.
-- **Text only.** DeepSeek rejects image, document and search-result
- content blocks in every format.
+- **Text only, from this CLI.** Since 2026-09-10 `deepseek-flash` takes
+ images through chat and Responses, and pro does not; the CLI has no
+ flag that sends one.
- **The Responses endpoint takes both models** since V4-Pro's official
release (2026-08-12); it refused pro before that.
- **FIM caps output at 4K tokens** and ignores thinking entirely.
diff --git a/gateway/DESIGN.md b/gateway/DESIGN.md
index ef9f854..a6ccfd5 100644
--- a/gateway/DESIGN.md
+++ b/gateway/DESIGN.md
@@ -281,25 +281,26 @@ scarce operation and gets the conservative boundary.
## Limits, and why these numbers
Worst-case cost per anonymous user per day, at flash off-peak rates
-($0.22/M input, $0.66/M output):
+($0.15/M input, $0.60/M output):
```
-60,000 input × $0.22/M = $0.0132
-20,000 output × $0.66/M = $0.0132
+60,000 input × $0.15/M = $0.0090
+20,000 output × $0.60/M = $0.0120
-------
- $0.0264 / user / day at full burn, off-peak
- $0.0528 / user / day at full burn, peak
+ $0.0210 / user / day at full burn, off-peak
+ $0.0420 / user / day at full burn, peak
```
-So a $1/day budget serves ~38 users burning *everything* off-peak, ~19 at
+So a $1/day budget serves ~47 users burning *everything* off-peak, ~23 at
peak, or several hundred normal ones — a normal turn is a few hundred
-input and a few hundred output tokens, about $0.0006. Roughly **1,700
-ordinary turns per dollar off-peak, 850 at peak.**
-
-These are the numbers of the card that took effect 2026-08-16 16:00 UTC,
-and they roughly halved the free tier's reach at an unchanged budget: the
-old flat card ($0.14/$0.28) made the same worst case $0.014/user/day. The
-quotas below have not been retuned for it — the budget breaker is what
+input and a few hundred output tokens, about $0.0005. Roughly **1,900
+ordinary turns per dollar off-peak, 950 at peak.**
+
+These are the numbers of the V4.1 Flash card of 2026-09-10. The card of
+2026-08-16 16:00 UTC ($0.22/$0.66) made the same worst case
+$0.0264/user/day, which had roughly halved the free tier's reach against
+the flat card before it ($0.14/$0.28, $0.014/user/day). The quotas below
+have not been retuned for either — the budget breaker is what
actually bounds spend, and it does its job at any card — but the reach
per dollar is now a time-of-day figure, and the peak windows are the
expensive seven hours.
diff --git a/gateway/internal/meter/meter.go b/gateway/internal/meter/meter.go
index bca50a7..9ac51e7 100644
--- a/gateway/internal/meter/meter.go
+++ b/gateway/internal/meter/meter.go
@@ -55,29 +55,57 @@ var RepriceAt = time.Date(2026, time.August, 16, 16, 0, 0, 0, time.UTC)
// The drift here runs the other way to the repricing's: over-charging our
// own budget is safe for the credit pool but it is still wrong, and it
// makes /economics report a cost-per-task the account never paid. The
-// same date gates the CLI's copy; `make price-check` guards the numbers,
-// and this comment is the only thing guarding the clock.
+// same date gates the CLI's copy; `make price-check` guards the numbers
+// and, since the V4.1 cut added a third instant, the dates too.
var weekendOffPeakAt = time.Date(2026, time.August, 22, 16, 0, 0, 0, time.UTC)
+// v41At is when the Flash card dropped with the DeepSeek-V4.1-Flash
+// release of 2026-09-10; Pro's card did not move. The instant is
+// INFERRED, as in the CLI's copy (internal/deepseek/pricing.go, V41At,
+// which records how): upstream published a date and no time. A cut
+// cannot drain the pool the way the repricing could, but it can make
+// /economics report a cost nobody paid.
+var v41At = time.Date(2026, time.September, 10, 11, 0, 0, 0, time.UTC)
+
// beijing is the vendor's clock; no daylight saving since 1991.
var beijing = time.FixedZone("CST", 8*60*60)
+// The cards are keyed by tier, not by model name. The API has accepted
+// three flash names (deepseek-v4-flash, deepseek-v4-flash-vision-exp,
+// deepseek-flash) and bills them all as flash, so cardFor maps a name
+// onto its tier and each card says what that tier cost in its era.
+const (
+ flashTier = "flash"
+ proTier = "pro"
+)
+
// ratesFlat is the card published 2026-08-02, in force before RepriceAt.
var ratesFlat = map[string]Price{
- "deepseek-v4-flash": {CacheHitInput: 0.0028, CacheMissInput: 0.14, Output: 0.28},
- "deepseek-v4-pro": {CacheHitInput: 0.003625, CacheMissInput: 0.435, Output: 0.87},
+ flashTier: {CacheHitInput: 0.0028, CacheMissInput: 0.14, Output: 0.28},
+ proTier: {CacheHitInput: 0.003625, CacheMissInput: 0.435, Output: 0.87},
+}
+
+// ratesV4OffPeak is the base card from RepriceAt until v41At; during
+// peakWindows every billing item costs peakMultiplier times these numbers.
+var ratesV4OffPeak = map[string]Price{
+ flashTier: {CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66},
+ proTier: {CacheHitInput: 0.022, CacheMissInput: 0.66, Output: 1.98},
+}
+
+// ratesV41OffPeak is the base card from v41At on: Flash cut on every
+// item, Pro unchanged. Same peak rule.
+var ratesV41OffPeak = map[string]Price{
+ flashTier: {CacheHitInput: 0.003, CacheMissInput: 0.15, Output: 0.6},
+ proTier: {CacheHitInput: 0.022, CacheMissInput: 0.66, Output: 1.98},
}
-// ratesOffPeak is the base card from RepriceAt on; during peakWindows
-// every billing item costs peakMultiplier times these numbers.
-var ratesOffPeak = map[string]Price{
- "deepseek-v4-flash": {CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66},
- // Released 2026-08-21, after the switchover, so it has no flat row
- // above. Priced identically to flash. Carried here even though the
- // policy allowlist does not admit it, so that a metering gap can never
- // be the reason it gets served for free.
- "deepseek-v4-flash-vision-exp": {CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66},
- "deepseek-v4-pro": {CacheHitInput: 0.022, CacheMissInput: 0.66, Output: 1.98},
+// offPeakCardAt is the time-of-day base card in force at t, which must
+// not be before RepriceAt.
+func offPeakCardAt(t time.Time) map[string]Price {
+ if t.Before(v41At) {
+ return ratesV4OffPeak
+ }
+ return ratesV41OffPeak
}
const peakMultiplier = 2.0
@@ -126,7 +154,7 @@ func PriceAt(model string, t time.Time) Price {
if t.Before(RepriceAt) {
return cardFor(ratesFlat, model)
}
- p := cardFor(ratesOffPeak, model)
+ p := cardFor(offPeakCardAt(t), model)
if inPeak(t) {
p = scale(p, peakMultiplier)
}
@@ -134,16 +162,13 @@ func PriceAt(model string, t time.Time) Price {
}
func cardFor(cards map[string]Price, model string) Price {
- if p, ok := cards[model]; ok {
- return p
- }
switch {
case strings.Contains(model, "pro") || model == "":
- return cards["deepseek-v4-pro"]
+ return cards[proTier]
case strings.Contains(model, "flash"):
- return cards["deepseek-v4-flash"]
+ return cards[flashTier]
default:
- return cards["deepseek-v4-pro"]
+ return cards[proTier]
}
}
@@ -228,10 +253,21 @@ func ceilingAt(model string, t time.Time) Price {
if t.Add(inFlight).Before(RepriceAt) {
return cardFor(ratesFlat, model)
}
+ // The base card is the admission instant's. The one change since
+ // the repricing, v41At, was a cut, so a request admitted before it is
+ // already reserved at the dearer card; a card that ever goes UP needs
+ // the same look-ahead the flip gets above.
if !t.Before(RepriceAt) && !peakTouches(t, inFlight) {
- return cardFor(ratesOffPeak, model)
+ return cardFor(offPeakCardAt(t), model)
+ }
+ return scale(cardFor(offPeakCardAt(maxTime(t, RepriceAt)), model), peakMultiplier)
+}
+
+func maxTime(a, b time.Time) time.Time {
+ if a.After(b) {
+ return a
}
- return scale(cardFor(ratesOffPeak, model), peakMultiplier)
+ return b
}
// peakTouches reports whether any instant of [t, t+d] falls in a peak
diff --git a/gateway/internal/meter/meter_test.go b/gateway/internal/meter/meter_test.go
index 7013667..bf5a0ae 100644
--- a/gateway/internal/meter/meter_test.go
+++ b/gateway/internal/meter/meter_test.go
@@ -233,7 +233,7 @@ func TestEstimateCeilingCoversTheNextPeriod(t *testing.T) {
// And far from any boundary, off-peak reserves at the off-peak card,
// not at a permanent doubling.
quiet := time.Date(2026, 8, 17, 12, 0, 0, 0, time.UTC)
- offPeak := costWith(cardFor(ratesOffPeak, model), Usage{InputTokens: 401, OutputTokens: 1000 + reasoningAllowance})
+ offPeak := costWith(cardFor(ratesV4OffPeak, model), Usage{InputTokens: 401, OutputTokens: 1000 + reasoningAllowance})
if est := EstimateAt(model, 400, 1000, false, quiet); math.Abs(est-offPeak) > 1e-12 {
t.Errorf("quiet off-peak reservation %v, want the off-peak card's %v", est, offPeak)
}
diff --git a/gateway/internal/meter/weekend_test.go b/gateway/internal/meter/weekend_test.go
index 6d31795..88dcd39 100644
--- a/gateway/internal/meter/weekend_test.go
+++ b/gateway/internal/meter/weekend_test.go
@@ -48,16 +48,42 @@ func TestWeekendIsReadOnTheVendorClock(t *testing.T) {
}
}
-func TestVisionVariantIsMetered(t *testing.T) {
+func TestEveryFlashNameIsMeteredAsFlash(t *testing.T) {
// A model priced upstream but missing from this card would be served
- // for free out of the budget pool.
- at := time.Date(2026, 8, 26, 12, 0, 0, 0, time.UTC)
- vision := PriceAt("deepseek-v4-flash-vision-exp", at)
- flash := PriceAt("deepseek-v4-flash", at)
- if vision != flash {
- t.Errorf("vision %+v != flash %+v", vision, flash)
- }
- if vision.CacheMissInput == 0 {
- t.Error("vision meters at zero")
+ // for free out of the budget pool, or, the other way, at pro rates
+ // that /economics then reports as spent. The API has billed three
+ // names as flash; each must land on the flash card of its instant.
+ for _, at := range []time.Time{
+ time.Date(2026, 8, 10, 12, 0, 0, 0, time.UTC), // flat
+ time.Date(2026, 8, 26, 12, 0, 0, 0, time.UTC), // V4, off-peak
+ time.Date(2026, 9, 14, 2, 0, 0, 0, time.UTC), // V4.1, peak
+ } {
+ flash := PriceAt("deepseek-v4-flash", at)
+ for _, name := range []string{"deepseek-flash", "deepseek-v4-flash-vision-exp", "deepseek-v4-flash-free"} {
+ if got := PriceAt(name, at); got != flash {
+ t.Errorf("%s at %v = %+v, want flash's %+v", name, at, got, flash)
+ }
+ }
+ if flash.CacheMissInput == 0 {
+ t.Errorf("flash meters at zero at %v", at)
+ }
+ }
+}
+
+func TestV41CutFlashAndLeftPro(t *testing.T) {
+ before := v41At.Add(-time.Minute) // 10:59 UTC Thursday, off-peak
+ if got := PriceAt("deepseek-flash", before); got != (Price{CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66}) {
+ t.Errorf("flash just before V4.1 = %+v, want the V4 card", got)
+ }
+ if got := PriceAt("deepseek-flash", v41At); got != (Price{CacheHitInput: 0.003, CacheMissInput: 0.15, Output: 0.6}) {
+ t.Errorf("flash at V4.1 = %+v, want the V4.1 card", got)
+ }
+ if a, b := PriceAt("deepseek-v4-pro", before), PriceAt("deepseek-v4-pro", v41At); a != b {
+ t.Errorf("pro moved at V4.1: %+v -> %+v", a, b)
+ }
+ // A reservation made before the cut must still cover a settlement
+ // after it: the ceiling may not be beaten by the clock.
+ if est, after := EstimateAt("deepseek-flash", 400, 1000, false, before), EstimateAt("deepseek-flash", 400, 1000, false, v41At); est < after {
+ t.Errorf("admitted before the cut the reservation %v is under the post-cut %v", est, after)
}
}
diff --git a/internal/cli/anthropic.go b/internal/cli/anthropic.go
index cb911cf..cff3de0 100644
--- a/internal/cli/anthropic.go
+++ b/internal/cli/anthropic.go
@@ -48,7 +48,7 @@ check how your DeepSeek key behaves for those tools before pointing them
at it.
Claude model names are accepted and remapped server-side: claude-opus* to
-deepseek-v4-pro, claude-sonnet*/claude-haiku* to deepseek-v4-flash, and
+deepseek-v4-pro, claude-sonnet*/claude-haiku* to deepseek-flash, and
anything unrecognised to flash. The usage line shows both names so the
cost is traceable to the model that actually ran.
diff --git a/internal/cli/chat.go b/internal/cli/chat.go
index 9ec8a67..82c6dd4 100644
--- a/internal/cli/chat.go
+++ b/internal/cli/chat.go
@@ -64,7 +64,7 @@ under reasoning_content.
}
fl := cmd.Flags()
- fl.StringVarP(&f.model, "model", "m", deepseek.ModelFlash, "model: deepseek-v4-flash or deepseek-v4-pro")
+ fl.StringVarP(&f.model, "model", "m", deepseek.ModelFlash, "model: deepseek-flash or deepseek-v4-pro")
fl.StringVarP(&f.system, "system", "s", "", "system prompt, or @file")
fl.StringVar(&f.think, "think", "", "thinking mode: on or off (default: the API's own default, on)")
fl.StringVarP(&f.effort, "effort", "e", "", "reasoning effort: low, high, max (pro promotes low to high)")
diff --git a/internal/cli/cli_test.go b/internal/cli/cli_test.go
index abe8650..4775a36 100644
--- a/internal/cli/cli_test.go
+++ b/internal/cli/cli_test.go
@@ -44,9 +44,16 @@ func TestMoneyKeepsSmallFiguresLegible(t *testing.T) {
func TestShortModelShowsAnthropicRemapping(t *testing.T) {
// When the endpoint silently swaps the model, the status line has to
// show both names or the cost looks unattributable.
- if got := shortModel("deepseek-v4-flash"); got != "flash" {
+ if got := shortModel(deepseek.ModelFlash); got != "flash" {
t.Errorf("got %q, want %q", got, "flash")
}
+ if got := shortModel(deepseek.ModelPro); got != "pro" {
+ t.Errorf("got %q, want %q", got, "pro")
+ }
+ // A retired name is served by its successor since 2026-09-10.
+ if got := shortModel("deepseek-v4-flash"); got != "deepseek-v4-flash→flash" {
+ t.Errorf("got %q, want both names", got)
+ }
if got := shortModel("claude-opus-4-1"); got != "claude-opus-4-1→pro" {
t.Errorf("got %q, want both names", got)
}
diff --git a/internal/cli/e2e_test.go b/internal/cli/e2e_test.go
index 3dee18a..a1d2704 100644
--- a/internal/cli/e2e_test.go
+++ b/internal/cli/e2e_test.go
@@ -405,18 +405,19 @@ func TestFIMSendsPromptAndSuffix(t *testing.T) {
}
func TestModelsJoinsThePublishedRateCard(t *testing.T) {
- got := runCLI(t, serve(`{"object":"list","data":[{"id":"deepseek-v4-flash","object":"model","owned_by":"deepseek"}]}`), "models")
+ got := runCLI(t, serve(`{"object":"list","data":[{"id":"deepseek-flash","object":"model","owned_by":"deepseek"}]}`), "models")
if got.err != nil {
t.Fatal(got.err)
}
- if !strings.Contains(got.stdout, "deepseek-v4-flash") {
+ if !strings.Contains(got.stdout, "deepseek-flash") {
t.Errorf("stdout = %q", got.stdout)
}
- // Which figure depends on the hour: flash cache-miss input is $0.22
- // off-peak and $0.44 peak. Both are published numbers, so asserting
- // on either keeps this a real check without making it a time bomb
- // that fails whenever the suite runs inside a peak window.
- if !strings.Contains(got.stdout, "0.22") && !strings.Contains(got.stdout, "0.44") {
+ // Which figure depends on the hour: flash cache-miss input is $0.15
+ // off-peak and $0.3 peak on the V4.1 card. Both are published
+ // numbers, so asserting on either keeps this a real check without
+ // making it a time bomb that fails whenever the suite runs inside a
+ // peak window.
+ if !strings.Contains(got.stdout, "$0.15") && !strings.Contains(got.stdout, "$0.3 ") {
t.Errorf("the price should sit next to the model, got %q", got.stdout)
}
}
@@ -490,7 +491,7 @@ func TestLedgerRecordsTheCallAndUsageReportsIt(t *testing.T) {
exec("chat", "hi", "--stream=false")
report := exec("usage", "--since", "all")
- if !strings.Contains(report, "deepseek-v4-flash") {
+ if !strings.Contains(report, "deepseek-flash") {
t.Errorf("usage did not report the call: %s", report)
}
// 64 of 100 prompt tokens came from cache.
diff --git a/internal/cli/fim.go b/internal/cli/fim.go
index cff5c2c..cb5eb54 100644
--- a/internal/cli/fim.go
+++ b/internal/cli/fim.go
@@ -48,7 +48,7 @@ non-thinking mode regardless of any effort setting.
}
fl := cmd.Flags()
- fl.StringVarP(&f.model, "model", "m", deepseek.ModelPro, "model: deepseek-v4-pro or deepseek-v4-flash")
+ fl.StringVarP(&f.model, "model", "m", deepseek.ModelPro, "model: deepseek-v4-pro or deepseek-flash")
fl.StringVar(&f.prefix, "prefix", "", "text before the gap, or @file (defaults to the positional argument or stdin)")
fl.StringVar(&f.suffix, "suffix", "", "text after the gap, or @file")
fl.IntVar(&f.maxTokens, "max-tokens", 0, "cap generated tokens (the endpoint's own ceiling is 4096)")
diff --git a/internal/cli/output.go b/internal/cli/output.go
index af4c3f9..a91b45c 100644
--- a/internal/cli/output.go
+++ b/internal/cli/output.go
@@ -152,15 +152,25 @@ func (o *Options) stats(api, requested string, u deepseek.Usage, dur time.Durati
// is always in --json and in the ledger; this is for the human glance.
func shortModel(model string) string {
resolved := deepseek.ResolveModel(model)
- short := strings.TrimPrefix(resolved, "deepseek-v4-")
+ short := trimModelPrefix(resolved)
if resolved != model {
- // The Anthropic endpoint remapped a Claude name; show both so the
- // cost is traceable to the model that actually ran.
+ // The Anthropic endpoint remapped a Claude name, or a retired name
+ // was served by its successor; show both so the cost is traceable
+ // to the model that actually ran.
return model + "→" + short
}
return short
}
+// trimModelPrefix drops the vendor prefix, and the generation where the
+// name carries one: deepseek-v4-pro reads pro, deepseek-flash reads flash.
+func trimModelPrefix(model string) string {
+ if short, ok := strings.CutPrefix(model, "deepseek-v4-"); ok {
+ return short
+ }
+ return strings.TrimPrefix(model, "deepseek-")
+}
+
// humanTokens renders a token count compactly: 940, 1.2k, 34k, 1.1M.
func humanTokens(n int) string {
switch {
diff --git a/internal/cli/pricing.go b/internal/cli/pricing.go
index 969aebc..ec569fa 100644
--- a/internal/cli/pricing.go
+++ b/internal/cli/pricing.go
@@ -34,11 +34,13 @@ type pricingResult struct {
NextChange string `json:"next_change"`
RepriceAt string `json:"reprice_at"`
+ CardSince string `json:"card_since"`
PeakWindowsUTC []string `json:"peak_windows_utc"`
PeakMultiplier float64 `json:"peak_multiplier"`
// Current is the card in effect at now_utc; OffPeak and Peak are the
- // dated cards that apply from reprice_at.
+ // full time-of-day card in force since card_since (before reprice_at,
+ // the dated card that is coming).
Current map[string]pricingPrice `json:"current"`
OffPeak map[string]pricingPrice `json:"off_peak"`
Peak map[string]pricingPrice `json:"peak"`
@@ -53,12 +55,14 @@ func newPricingCmd(o *Options) *cobra.Command {
Long: strings.TrimSpace(`
Answer what a token costs at this instant, and when that changes.
-DeepSeek's repricing of 2026-08-13 is dated: until 16:00 UTC on
-2026-08-16 every hour bills at the flat card of 2026-08-02, and from
-that instant billing is peak/off-peak on a new, higher card — peak hours
-01:00-04:00 and 06:00-10:00 UTC on weekdays at twice the off-peak
-rate. Since 2026-08-22 weekends bill off-peak all day, on the Beijing
-calendar, so peak is 35 hours a week rather than 49.
+Billing has been peak/off-peak since 16:00 UTC on 2026-08-16 (before
+it, every hour billed at the flat card of 2026-08-02): peak hours
+01:00-04:00 and 06:00-10:00 UTC on weekdays, at twice the off-peak rate.
+Since 2026-08-22 weekends bill off-peak all day, on the Beijing
+calendar, so peak is 35 hours a week rather than 49. Since the V4.1
+Flash release of 2026-09-10 the Flash card is lower; Pro's is unchanged.
+The retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp
+bill as deepseek-flash.
This command reads no network and spends nothing: the schedule is the
same one the cost estimates use, so what it prints is what the usage
@@ -82,6 +86,7 @@ func pricingAt(now time.Time) *pricingResult {
Multiplier: period.Multiplier,
NextChange: deepseek.NextChange(now).Format(time.RFC3339),
RepriceAt: deepseek.RepriceAt.Format(time.RFC3339),
+ CardSince: deepseek.CardSince(fullCardAt(now)).Format(time.RFC3339),
PeakMultiplier: deepseek.PeakMultiplier,
Current: map[string]pricingPrice{},
OffPeak: map[string]pricingPrice{},
@@ -96,7 +101,7 @@ func pricingAt(now time.Time) *pricingResult {
if p, ok := deepseek.PriceAt(m, now); ok {
res.Current[m] = pricingPrice{p.CacheHitInput, p.CacheMissInput, p.Output}
}
- if p, ok := deepseek.PriceAt(m, deepseek.RepriceAt); ok {
+ if p, ok := deepseek.BasePriceAt(m, fullCardAt(now)); ok {
res.OffPeak[m] = pricingPrice{p.CacheHitInput, p.CacheMissInput, p.Output}
res.Peak[m] = pricingPrice{
p.CacheHitInput * deepseek.PeakMultiplier,
@@ -146,17 +151,18 @@ func formatPricing(now time.Time) string {
for _, win := range deepseek.PeakWindows {
windows = append(windows, fmtMinutes(win.Start)+"-"+fmtMinutes(win.End))
}
+ since := deepseek.CardSince(fullCardAt(now)).Format("2006-01-02 15:04 UTC")
if now.Before(deepseek.RepriceAt) {
fmt.Fprintf(&b, "\nfrom %s — peak hours %s UTC, Mon-Fri; all other hours, and all weekend, off-peak at half of peak:\n",
- deepseek.RepriceAt.Format("2006-01-02 15:04 UTC"), strings.Join(windows, " and "))
+ since, strings.Join(windows, " and "))
} else {
- fmt.Fprintf(&b, "\nthe full card — peak hours %s UTC, Mon-Fri; all other hours, and all weekend, off-peak at half of peak:\n",
- strings.Join(windows, " and "))
+ fmt.Fprintf(&b, "\nthe full card, since %s — peak hours %s UTC, Mon-Fri; all other hours, and all weekend, off-peak at half of peak:\n",
+ since, strings.Join(windows, " and "))
}
w = tabwriter.NewWriter(&b, 0, 0, 2, ' ', 0)
fmt.Fprintln(w, "MODEL\tPERIOD\tIN (CACHED)\tIN (MISS)\tOUT")
for _, m := range deepseek.Models {
- p, ok := deepseek.PriceAt(m, deepseek.RepriceAt)
+ p, ok := deepseek.BasePriceAt(m, fullCardAt(now))
if !ok {
continue
}
@@ -166,10 +172,21 @@ func formatPricing(now time.Time) string {
}
w.Flush()
+ fmt.Fprintf(&b, "\nthe retired names %s and %s bill as %s.", deepseek.ModelFlashV4, deepseek.ModelFlashVision, deepseek.ModelFlash)
fmt.Fprint(&b, "\ncost estimates switch cards on the effective instant automatically.\nupstream copy, offline: deepseek docs show quick_start/pricing")
return b.String()
}
+// fullCardAt is the instant whose era supplies the full time-of-day card
+// shown beside the current price: now, or the repricing instant while the
+// flat card is still in force, so a reader sees the card that is coming.
+func fullCardAt(now time.Time) time.Time {
+ if now.Before(deepseek.RepriceAt) {
+ return deepseek.RepriceAt
+ }
+ return now
+}
+
func fmtMinutes(m int) string {
return fmt.Sprintf("%02d:%02d", m/60, m%60)
}
diff --git a/internal/cli/pricing_test.go b/internal/cli/pricing_test.go
index a6a44a0..40cb5a7 100644
--- a/internal/cli/pricing_test.go
+++ b/internal/cli/pricing_test.go
@@ -22,15 +22,15 @@ func TestPricingBeforeTheFlip(t *testing.T) {
if res.NextChange != deepseek.RepriceAt.Format(time.RFC3339) {
t.Errorf("next change = %s, want the repricing instant", res.NextChange)
}
- if got := res.Current["deepseek-v4-flash"]; got.CacheMissInput != 0.14 {
+ if got := res.Current[deepseek.ModelFlash]; got.CacheMissInput != 0.14 {
t.Errorf("current flash miss = %v, want the flat 0.14", got.CacheMissInput)
}
// The dated card rides along so a script can see the future without
// a second source.
- if got := res.OffPeak["deepseek-v4-pro"]; got.Output != 1.98 {
+ if got := res.OffPeak[deepseek.ModelPro]; got.Output != 1.98 {
t.Errorf("off-peak pro output = %v, want 1.98", got.Output)
}
- if got, want := res.Peak["deepseek-v4-flash"].CacheMissInput, 0.44; got != want {
+ if got, want := res.Peak[deepseek.ModelFlash].CacheMissInput, 0.44; got != want {
t.Errorf("peak flash miss = %v, want %v (2x off-peak)", got, want)
}
@@ -49,7 +49,7 @@ func TestPricingInsideAPeakWindow(t *testing.T) {
if res.Period != "peak" || res.Multiplier != deepseek.PeakMultiplier {
t.Errorf("period = %s at %gx, want peak at %gx", res.Period, res.Multiplier, deepseek.PeakMultiplier)
}
- if got := res.Current["deepseek-v4-flash"]; got.CacheMissInput != 0.44 {
+ if got := res.Current[deepseek.ModelFlash]; got.CacheMissInput != 0.44 {
t.Errorf("current flash miss = %v, want the peak 0.44", got.CacheMissInput)
}
if res.NextChange != time.Date(2026, 8, 17, 4, 0, 0, 0, time.UTC).Format(time.RFC3339) {
@@ -62,6 +62,36 @@ func TestPricingInsideAPeakWindow(t *testing.T) {
}
}
+func TestPricingAfterV41ShowsTheNewFlashCard(t *testing.T) {
+ now := time.Date(2026, 9, 17, 12, 0, 0, 0, time.UTC) // Thursday, off-peak
+ res := pricingAt(now)
+
+ if res.CardSince != deepseek.V41At.Format(time.RFC3339) {
+ t.Errorf("card_since = %s, want the V4.1 instant", res.CardSince)
+ }
+ // The full card follows the era in force, not the repricing instant:
+ // reading it at RepriceAt showed the V4 card for a week after it died.
+ if got := res.OffPeak[deepseek.ModelFlash]; got.CacheMissInput != 0.15 {
+ t.Errorf("off-peak flash miss = %v, want the V4.1 0.15", got.CacheMissInput)
+ }
+ if got := res.Peak[deepseek.ModelFlash]; got.Output != 1.2 {
+ t.Errorf("peak flash output = %v, want 1.2", got.Output)
+ }
+ if got := res.Peak[deepseek.ModelPro]; got.Output != 3.96 {
+ t.Errorf("peak pro output = %v, want the unchanged 3.96", got.Output)
+ }
+ if _, listed := res.Current[deepseek.ModelFlashV4]; listed {
+ t.Error("a retired name has its own row; it should bill as deepseek-flash")
+ }
+
+ text := formatPricing(now)
+ for _, want := range []string{"since 2026-09-10 11:00 UTC", "$0.15", "$0.6", "deepseek-v4-flash-vision-exp bill as deepseek-flash"} {
+ if !strings.Contains(text, want) {
+ t.Errorf("text output is missing %q:\n%s", want, text)
+ }
+ }
+}
+
func TestPricingJSONAgreesWithTheCostEstimator(t *testing.T) {
// The command and Cost() must read the same table: the card `pricing`
// prints for an instant is the card a call at that instant is priced
@@ -70,6 +100,8 @@ func TestPricingJSONAgreesWithTheCostEstimator(t *testing.T) {
time.Date(2026, 8, 14, 12, 0, 0, 0, time.UTC),
time.Date(2026, 8, 17, 2, 0, 0, 0, time.UTC),
time.Date(2026, 8, 17, 12, 0, 0, 0, time.UTC),
+ time.Date(2026, 9, 14, 2, 0, 0, 0, time.UTC),
+ time.Date(2026, 9, 17, 12, 0, 0, 0, time.UTC),
} {
res := pricingAt(now)
for _, m := range []string{deepseek.ModelFlash, deepseek.ModelPro} {
diff --git a/internal/cli/respond.go b/internal/cli/respond.go
index 53f6652..9ef4ece 100644
--- a/internal/cli/respond.go
+++ b/internal/cli/respond.go
@@ -55,7 +55,7 @@ Both models are accepted since V4-Pro's official release
}
fl := cmd.Flags()
- fl.StringVarP(&f.model, "model", "m", deepseek.ModelFlash, "model: deepseek-v4-flash or deepseek-v4-pro")
+ fl.StringVarP(&f.model, "model", "m", deepseek.ModelFlash, "model: deepseek-flash or deepseek-v4-pro")
fl.StringVarP(&f.instructions, "instructions", "s", "", "system-level instructions, or @file")
fl.StringVarP(&f.effort, "effort", "e", "", "reasoning effort: none, low, high, or max (none disables thinking)")
fl.IntVar(&f.maxTokens, "max-tokens", 0, "cap generated tokens, reasoning included")
diff --git a/internal/cli/status.go b/internal/cli/status.go
index e5b25f0..81766ac 100644
--- a/internal/cli/status.go
+++ b/internal/cli/status.go
@@ -116,7 +116,7 @@ func formatStatus(res *statusResult) string {
func shortModels(ids []string) []string {
out := make([]string, len(ids))
for i, id := range ids {
- out[i] = strings.TrimPrefix(id, "deepseek-v4-")
+ out[i] = trimModelPrefix(id)
}
return out
}
diff --git a/internal/deepseek/client.go b/internal/deepseek/client.go
index 75b8214..0521ed9 100644
--- a/internal/deepseek/client.go
+++ b/internal/deepseek/client.go
@@ -26,13 +26,25 @@ import (
const DefaultBaseURL = "https://api.deepseek.com"
// Models. DeepSeek exposes exactly two; everything else the API accepts
-// (Claude model names on the Anthropic endpoint) is mapped server-side.
+// (Claude model names on the Anthropic endpoint, retired model names) is
+// mapped server-side.
+//
+// ModelFlash is a floating alias: since 2026-09-10 it serves
+// DeepSeek-V4.1-Flash, and whichever Flash comes next will answer to it
+// too. That is the name upstream tells callers to use.
const (
- ModelFlash = "deepseek-v4-flash"
+ ModelFlash = "deepseek-flash"
ModelPro = "deepseek-v4-pro"
- // ModelFlashVision is the experimental multimodal variant released
- // 2026-08-21. It takes image input and bills at exactly the Flash
- // rates; it does not support FIM completion.
+)
+
+// Retired names. The models behind them were retired on 2026-09-10; the
+// API still accepts both, serves them with V4.1 Flash and bills them at
+// the Flash price, so they resolve to ModelFlash. Kept as constants
+// because ledger rows written before that date carry them.
+const (
+ ModelFlashV4 = "deepseek-v4-flash"
+ // ModelFlashVision was the experimental multimodal variant of
+ // 2026-08-21. V4.1 Flash takes images natively.
ModelFlashVision = "deepseek-v4-flash-vision-exp"
)
diff --git a/internal/deepseek/playground_card_test.go b/internal/deepseek/playground_card_test.go
new file mode 100644
index 0000000..327ef7f
--- /dev/null
+++ b/internal/deepseek/playground_card_test.go
@@ -0,0 +1,39 @@
+package deepseek
+
+import (
+ "os"
+ "regexp"
+ "strconv"
+ "testing"
+ "time"
+)
+
+// The playground prices calls in the browser from its own copy of the
+// Flash card, because the site is static and cannot ask the CLI. That copy
+// went unchecked through two changes: it read the dead card for a day
+// after the 2026-08-16 repricing, and it billed weekends at peak for
+// weeks after 2026-08-22. So it is pinned here to the newest Flash card
+// this package encodes.
+//
+// Newest, not current: reading the card at time.Now() would turn red on
+// the day a dated card takes effect, whoever happened to push that day.
+func TestPlaygroundCarriesTheNewestFlashCard(t *testing.T) {
+ src, err := os.ReadFile("../../site/playground.js")
+ if err != nil {
+ t.Fatal(err)
+ }
+ m := regexp.MustCompile(`var RATES = \{ cacheHit: ([0-9.]+), cacheMiss: ([0-9.]+), output: ([0-9.]+) \};`).FindSubmatch(src)
+ if m == nil {
+ t.Fatal("site/playground.js: no RATES line; if it moved, move this test with it")
+ }
+ var got [3]float64
+ for i := range got {
+ if got[i], err = strconv.ParseFloat(string(m[i+1]), 64); err != nil {
+ t.Fatal(err)
+ }
+ }
+ want, _ := BasePriceAt(ModelFlash, time.Date(9999, 1, 1, 0, 0, 0, 0, time.UTC))
+ if got != [3]float64{want.CacheHitInput, want.CacheMissInput, want.Output} {
+ t.Errorf("playground RATES = %v, newest flash card = %+v", got, want)
+ }
+}
diff --git a/internal/deepseek/pricing.go b/internal/deepseek/pricing.go
index 9e1c690..57c2282 100644
--- a/internal/deepseek/pricing.go
+++ b/internal/deepseek/pricing.go
@@ -72,34 +72,84 @@ var RepriceAt = time.Date(2026, time.August, 16, 16, 0, 0, 0, time.UTC)
// 2026-08-17 or Saturday 2026-08-22 really did bill peak.
var WeekendOffPeakAt = time.Date(2026, time.August, 22, 16, 0, 0, 0, time.UTC)
+// V41At is when the Flash card dropped with the DeepSeek-V4.1-Flash
+// release. The changelog dates the release 2026-09-10 and says only that
+// "API prices have been reduced accordingly"; the pricing page carries
+// the new card and no effective instant. Pro's card did not move.
+//
+// The instant is INFERRED, not published. Our docs mirror fetched the
+// pricing page at 04:50 UTC on 2026-09-10 and got the old card, and at
+// 11:27 UTC the same day and got the new one. 11:00 UTC is the last whole
+// hour inside that bracket, which errs toward the old, dearer card: an
+// estimate for a call in the bracket can overstate what it cost, never
+// understate it. Move it if DeepSeek publishes the real instant.
+var V41At = time.Date(2026, time.September, 10, 11, 0, 0, 0, time.UTC)
+
// beijing is the vendor's clock. China has observed no daylight saving
// since 1991, so a fixed offset is exact and needs no tzdata.
var beijing = time.FixedZone("CST", 8*60*60)
+// The cards are keyed by ModelFlash and ModelPro as the two tiers, and
+// ResolveModel maps every name the API accepts onto one of them. Each
+// card is what that tier cost in its era, so ModelFlash's flat row is
+// what deepseek-v4-flash cost then.
+//
+// Superseded cards stay because the ledger stores token counts rather
+// than dollars, and a call must reprice under the card it was actually
+// billed at.
+
// pricesFlat is the card published 2026-08-02, in force before RepriceAt.
-// That instant has passed, so nothing live prices against it any more; it
-// stays because the ledger stores token counts rather than dollars, and a
-// call made before the flip must still reprice under the card it was
-// actually billed at.
var pricesFlat = map[string]Price{
ModelFlash: {CacheHitInput: 0.0028, CacheMissInput: 0.14, Output: 0.28},
ModelPro: {CacheHitInput: 0.003625, CacheMissInput: 0.435, Output: 0.87},
}
-// pricesOffPeak is the base card from RepriceAt on. During PeakWindows
-// every billing item costs PeakMultiplier times these numbers; DeepSeek
-// publishes the peak figures rather than the rule, and they are exactly
-// double, so the multiplier is data, not interpretation.
-var pricesOffPeak = map[string]Price{
- ModelFlash: {CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66},
- ModelFlashVision: {CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66},
- ModelPro: {CacheHitInput: 0.022, CacheMissInput: 0.66, Output: 1.98},
+// pricesV4OffPeak is the base card from RepriceAt until V41At. During
+// PeakWindows every billing item costs PeakMultiplier times these
+// numbers; DeepSeek publishes the peak figures rather than the rule, and
+// they are exactly double, so the multiplier is data, not interpretation.
+var pricesV4OffPeak = map[string]Price{
+ ModelFlash: {CacheHitInput: 0.007, CacheMissInput: 0.22, Output: 0.66},
+ ModelPro: {CacheHitInput: 0.022, CacheMissInput: 0.66, Output: 1.98},
+}
+
+// pricesV41OffPeak is the base card from V41At on: Flash cut on every
+// item, Pro unchanged. Same peak rule.
+var pricesV41OffPeak = map[string]Price{
+ ModelFlash: {CacheHitInput: 0.003, CacheMissInput: 0.15, Output: 0.6},
+ ModelPro: {CacheHitInput: 0.022, CacheMissInput: 0.66, Output: 1.98},
}
// Models are the priced models, in the order the rate card lists them.
-// One list so a new model reaches every table at once: the vision variant
-// was published upstream and absent here, which meters it at zero.
-var Models = []string{ModelFlash, ModelFlashVision, ModelPro}
+// One list so a new model reaches every table at once. The retired names
+// are not listed: they bill as ModelFlash, and the card says so once.
+var Models = []string{ModelFlash, ModelPro}
+
+// cardAt is the base card of the era in force at t.
+func cardAt(t time.Time) map[string]Price {
+ switch {
+ case t.Before(RepriceAt):
+ return pricesFlat
+ case t.Before(V41At):
+ return pricesV4OffPeak
+ default:
+ return pricesV41OffPeak
+ }
+}
+
+// CardSince is the instant the base card in force at t took effect, and
+// the zero time for the flat card, which has no start this schedule
+// knows of.
+func CardSince(t time.Time) time.Time {
+ switch {
+ case t.Before(RepriceAt):
+ return time.Time{}
+ case t.Before(V41At):
+ return RepriceAt
+ default:
+ return V41At
+ }
+}
// PeakMultiplier scales the off-peak card during PeakWindows.
const PeakMultiplier = 2.0
@@ -160,7 +210,7 @@ func isBeijingWeekend(t time.Time) bool {
// NextChange is the next instant after t at which the price of a call
// changes: the repricing instant while the flat card is in force, then
-// the nearest peak-window boundary of the UTC day.
+// the nearest peak-window boundary or card change.
func NextChange(t time.Time) time.Time {
if t.Before(RepriceAt) {
return RepriceAt
@@ -169,12 +219,13 @@ func NextChange(t time.Time) time.Time {
// turnover at 16:00 UTC, and the policy start dates — lands on a UTC
// hour, so walking hours finds the next change exactly. Scanning a
// week of them covers the longest run of one period, Friday 10:00
- // UTC to Monday 01:00 UTC, with room to spare.
+ // UTC to Monday 01:00 UTC, with room to spare. A new card is a change
+ // even when the period's label is not.
u := t.UTC()
- here := PeriodAt(u).Label
+ here, since := PeriodAt(u).Label, CardSince(u)
at := u.Truncate(time.Hour).Add(time.Hour)
for i := 0; i < 8*24; i++ {
- if PeriodAt(at).Label != here {
+ if PeriodAt(at).Label != here || !CardSince(at).Equal(since) {
return at
}
at = at.Add(time.Hour)
@@ -182,14 +233,18 @@ func NextChange(t time.Time) time.Time {
return at
}
+// BasePriceAt is a model's base card in the era in force at t, before
+// any peak multiplier: the off-peak row once time-of-day billing began,
+// the flat row before it.
+func BasePriceAt(model string, t time.Time) (Price, bool) {
+ p, ok := cardAt(t)[ResolveModel(model)]
+ return p, ok
+}
+
// PriceAt returns the effective rate card for a model at one instant:
// that era's base card, scaled by the period's multiplier.
func PriceAt(model string, t time.Time) (Price, bool) {
- cards := pricesFlat
- if !t.Before(RepriceAt) {
- cards = pricesOffPeak
- }
- p, ok := cards[ResolveModel(model)]
+ p, ok := BasePriceAt(model, t)
if !ok {
return Price{}, false
}
@@ -213,11 +268,14 @@ func PriceFor(model string) (Price, bool) {
//
// The Anthropic-format endpoint accepts Claude model names and remaps
// them: claude-opus* becomes pro, claude-haiku*/claude-sonnet* become
-// flash, and anything else unrecognised falls back to flash.
+// flash, and anything else unrecognised falls back to flash. The retired
+// V4 Flash names are served and billed as flash.
func ResolveModel(model string) string {
switch {
case model == ModelFlash || model == ModelPro:
return model
+ case model == ModelFlashV4 || model == ModelFlashVision:
+ return ModelFlash
case hasPrefix(model, "claude-opus"):
return ModelPro
case hasPrefix(model, "claude-haiku"), hasPrefix(model, "claude-sonnet"):
diff --git a/internal/deepseek/usage_test.go b/internal/deepseek/usage_test.go
index 5b72bf7..2f080d1 100644
--- a/internal/deepseek/usage_test.go
+++ b/internal/deepseek/usage_test.go
@@ -294,17 +294,65 @@ func TestNextChangeCrossesTheWeekend(t *testing.T) {
}
}
-func TestVisionModelIsPricedAtFlashRates(t *testing.T) {
- // Released 2026-08-21, priced identically to Flash on every bucket in
- // both published currencies. Absent from the card it would meter at zero.
- at := time.Date(2026, 8, 26, 12, 0, 0, 0, time.UTC) // Wednesday, off-peak
- vision, ok := PriceAt(ModelFlashVision, at)
- if !ok {
- t.Fatal("deepseek-v4-flash-vision-exp has no published rate")
+func TestV41FlashCardSwitchesOnItsInstant(t *testing.T) {
+ // 2026-09-10: the Flash card dropped with V4.1, Pro's did not. 11:00
+ // UTC is off-peak on a Thursday, so the flip lands on the off-peak
+ // rows of both cards.
+ u := Usage{InputTokens: 2_000_000, CacheHitTokens: 1_000_000, CacheMissTokens: 1_000_000, OutputTokens: 1_000_000}
+
+ before, _ := CostAt(ModelFlash, u, V41At.Add(-time.Nanosecond))
+ if want := 0.007 + 0.22 + 0.66; math.Abs(before-want) > 1e-9 {
+ t.Errorf("one instant before V4.1: got %v, want the V4 card's %v", before, want)
+ }
+ after, _ := CostAt(ModelFlash, u, V41At)
+ if want := 0.003 + 0.15 + 0.6; math.Abs(after-want) > 1e-9 {
+ t.Errorf("at V4.1: got %v, want the V4.1 card's %v", after, want)
+ }
+
+ // Peak is still exactly double, on the new card.
+ peak := time.Date(2026, 9, 14, 2, 0, 0, 0, time.UTC) // Monday, 01:00-04:00 UTC
+ p, _ := PriceAt(ModelFlash, peak)
+ if (p != Price{CacheHitInput: 0.006, CacheMissInput: 0.3, Output: 1.2}) {
+ t.Errorf("flash at peak on the V4.1 card = %+v, want 0.006/0.3/1.2", p)
+ }
+
+ for _, at := range []time.Time{V41At.Add(-time.Nanosecond), V41At, peak} {
+ pro, _ := CostAt(ModelPro, Usage{OutputTokens: 1_000_000}, at)
+ want := 1.98 * PeriodAt(at).Multiplier
+ if math.Abs(pro-want) > 1e-9 {
+ t.Errorf("pro output at %v = %v, want %v: V4.1 did not reprice pro", at, pro, want)
+ }
+ }
+}
+
+func TestRetiredFlashNamesBillAsFlashInEveryEra(t *testing.T) {
+ // deepseek-v4-flash and deepseek-v4-flash-vision-exp are served by
+ // V4.1 Flash since 2026-09-10 and billed at the Flash price; before it
+ // they WERE flash (vision priced identically from its release). Either
+ // way the right answer is the flash card of the instant.
+ for _, at := range []time.Time{atFlat, atOffPeak, atPeak, V41At, V41At.Add(24 * time.Hour)} {
+ flash, _ := PriceAt(ModelFlash, at)
+ for _, name := range []string{ModelFlashV4, ModelFlashVision} {
+ got, ok := PriceAt(name, at)
+ if !ok || got != flash {
+ t.Errorf("%s at %v = %+v (ok=%v), want flash's %+v", name, at, got, ok, flash)
+ }
+ }
+ }
+}
+
+func TestNextChangeSeesTheCardChange(t *testing.T) {
+ // 10:00 to 11:00 UTC on 2026-09-10 is off-peak on both sides of the
+ // flip, so only the card changes; a label-only walk would sail past it.
+ at := V41At.Add(-30 * time.Minute)
+ if got := NextChange(at); !got.Equal(V41At) {
+ t.Errorf("NextChange(%v) = %v, want the card change at %v", at, got, V41At)
+ }
+ if got := CardSince(V41At.Add(time.Hour)); !got.Equal(V41At) {
+ t.Errorf("CardSince after V4.1 = %v, want %v", got, V41At)
}
- flash, _ := PriceAt(ModelFlash, at)
- if vision != flash {
- t.Errorf("vision %+v != flash %+v", vision, flash)
+ if got := CardSince(atOffPeak); !got.Equal(RepriceAt) {
+ t.Errorf("CardSince in the V4 era = %v, want %v", got, RepriceAt)
}
}
@@ -333,6 +381,8 @@ func TestResolveModel(t *testing.T) {
cases := map[string]string{
ModelFlash: ModelFlash,
ModelPro: ModelPro,
+ ModelFlashV4: ModelFlash, // retired 2026-09-10, served by V4.1 Flash
+ ModelFlashVision: ModelFlash,
"claude-opus-4-1": ModelPro,
"claude-sonnet-4-5": ModelFlash,
"claude-haiku-4-5": ModelFlash,
diff --git a/internal/docs/corpus.tar.gz b/internal/docs/corpus.tar.gz
index e0d5972..1b2bb90 100644
Binary files a/internal/docs/corpus.tar.gz and b/internal/docs/corpus.tar.gz differ
diff --git a/site/bench/index.html b/site/bench/index.html
index 5c95a90..1097dfd 100644
--- a/site/bench/index.html
+++ b/site/bench/index.html
@@ -176,16 +176,17 @@ | Per 1M tokens | v4-flash | v4-pro | GPT-5.6 Sol | pro is cheaper by |
|---|---|---|---|---|
| Per 1M tokens | flash | v4-pro | GPT-5.6 Sol | pro is cheaper by |
| input, cache miss | $0.22 / $0.44 | $0.66 / $1.32 | $5.00 | 7.6x / 3.8x |
| input, cache hit | $0.007 / $0.014 | $0.022 / $0.044 | $0.50 | 23x / 11x |
| output | $0.66 / $1.32 | $1.98 / $3.96 | $30.00 | 15x / 7.6x |
| input, cache miss | $0.15 / $0.30 | $0.66 / $1.32 | $5.00 | 7.6x / 3.8x |
| input, cache hit | $0.003 / $0.006 | $0.022 / $0.044 | $0.50 | 23x / 11x |
| output | $0.60 / $1.20 | $1.98 / $3.96 | $30.00 | 15x / 7.6x |
DeepSeek cells read off-peak / peak, on the -card in force since 2026-08-16 16:00 UTC; they are a conversion of the RMB +
DeepSeek cells read off-peak / peak: pro on the
+card in force since 2026-08-16 16:00 UTC, flash (deepseek-flash,
+V4.1) on the lower card of 2026-09-10. They are a conversion of the RMB
card (pro: ¥4.5 / ¥0.15 / ¥13.5 per 1M off-peak, double at peak)
at one consistent rate. GPT-5.6 Sol prices are from OpenAI's own listing.
The pricing page has the full schedule.
-
+
https://api.deepseek.com
-ok GET /models 141ms deepseek-v4-flash, deepseek-v4-pro
+ok GET /models 141ms deepseek-flash, deepseek-v4-pro
ok GET /user/balance 126ms 18.48 CNY
ok POST /chat/completions 651ms 5 in / 1 out
ok POST /anthropic/v1/messages 590ms 5 in / 1 out
@@ -744,9 +744,10 @@ def jstr(s):
Costs are estimates. Computed from DeepSeek's published
USD rate card, not from your invoice. Token counts are exact, and they are what
gets stored, so old calls can be repriced when the card changes.
-Text only. DeepSeek rejects image, document and
-search-result content blocks in every format – that is the API, not a
-gap here.
+Text only. The CLI sends text. Since 2026-09-10
+deepseek-flash also takes images through chat completions and
+Responses, and deepseek-v4-pro does not; the CLI has no flag
+that sends one.
""",
))
@@ -874,7 +875,7 @@ def jstr(s):
Flag Effect -m, --modeldeepseek-v4-flash (default) or deepseek-v4-pro-m, --modeldeepseek-flash (default) or deepseek-v4-pro-s, --systemSystem prompt, inline or @file --think on|offThinking mode. Default is the API's own, which is on -e, --effortlow, high or maxFour wire formats
DeepSeek exposes the same two models –
-deepseek-v4-flash and deepseek-v4-pro – through
+deepseek-flash and deepseek-v4-pro – through
four different request shapes, so that existing ecosystems can point at it
without code changes. They are not interchangeable.
@@ -1184,8 +1185,8 @@ def jstr(s):
You send You get claude-opus-*deepseek-v4-proclaude-sonnet-*, claude-haiku-*deepseek-v4-flashanything unrecognised deepseek-v4-flashclaude-sonnet-*, claude-haiku-*deepseek-flashanything unrecognised deepseek-flash
deepseek-flash takes images through chat completions and
+Responses; deepseek-v4-pro is text only.--effort: +79 input tokens on flash at the default,
-+92 at max, and nothing at all at
-low – where the model still reasons.
+depends on --effort: on V4 flash it was +79 input tokens at the
+default, +92 at max, and nothing at all at
+low – where the model still reasons. V4.1 Flash has not
+been re-measured.
The measured table.: keep-alive comments for up to ten minutes before inference
@@ -1229,13 +1232,13 @@ def jstr(s):
slug="cost/",
crumb="cost",
title="What DeepSeek actually costs: cache math and a local usage ledger",
- description="DeepSeek's context cache makes a cached input token 50x cheaper than an uncached one. How the deepseek CLI prices every call, what the local usage ledger records, and the caveats on every figure it prints.",
- keywords="deepseek pricing, deepseek api cost, deepseek context cache, deepseek cache hit tokens, deepseek token cost calculator, deepseek v4 flash price, deepseek usage tracking",
+ description="DeepSeek's context cache makes a cached input token 50x cheaper than an uncached one on flash and 30x on pro. How the deepseek CLI prices every call, what the local usage ledger records, and the caveats on every figure it prints.",
+ keywords="deepseek pricing, deepseek api cost, deepseek context cache, deepseek cache hit tokens, deepseek token cost calculator, deepseek flash price, deepseek v4.1 flash price, deepseek usage tracking",
jsonld=tech_article("What DeepSeek actually costs", "Pricing, context-cache savings, and the deepseek CLI usage ledger.", "cost/"),
body="""
DeepSeek's headline feature is a disk-backed context cache that -makes a repeated prompt prefix roughly thirty times cheaper. Since +makes a repeated prompt prefix thirty to fifty times cheaper. Since 2026-08-16 a second lever sits on top of it: the hour of the day, worth 2×. Neither is visible unless something is counting – so this counts.
@@ -1249,22 +1252,23 @@ def jstr(s):| Model | Period | Input (cached) | Input (miss) | Output |
|---|---|---|---|---|
deepseek-v4-flash | off-peak | $0.007 | $0.22 | $0.66 |
| peak | $0.014 | $0.44 | $1.32 | |
deepseek-v4-flash-vision-exp | off-peak | $0.007 | $0.22 | $0.66 |
| peak | $0.014 | $0.44 | $1.32 | |
deepseek-flash | off-peak | $0.003 | $0.15 | $0.60 |
| peak | $0.006 | $0.30 | $1.20 | |
deepseek-v4-pro | off-peak | $0.022 | $0.66 | $1.98 |
| peak | $0.044 | $1.32 | $3.96 |
ds models prints the card in force next to the live model
+
The retired names deepseek-v4-flash and
+deepseek-v4-flash-vision-exp still work: V4.1 Flash serves them,
+at the deepseek-flash price. ds models prints the card in force next to the live model
list, so the price is on screen when you pick, and ds pricing
prints the full schedule with the period you are in right now.
On flash, a cache hit costs 1/31st of a miss per token. -The same 2,900-token prompt, sent twice:
+On flash, a cache hit costs 1/50th of a miss per token; +on pro, 1/30th. The same 2,900-token prompt, sent twice, on the V4 flash +card of the day, where the ratio was 1/31st:
$ ds chat "..." --system @prefix.txt
@@ -1279,7 +1283,8 @@ def jstr(s):
same prefix again. It is 15× and not 31× because the 3% that
changed – the tail of the prompt, and the question itself –
still bills at the miss rate: the realised saving is always the rate ratio
-times how much of the prompt you kept identical. Hence the practical rule:
+times how much of the prompt you kept identical. On today's flash card the
+same call drops about 20×. Hence the practical rule:
put the stable part of a prompt first – same system
prompt, same files, in the same order – and let the variable part come
last.
@@ -1305,7 +1310,7 @@ def jstr(s):
---effortflash pro thinking --effortV4 flash pro thinking none+0 +0 off entirely minimal, low+0 +0 on
+The flash column is V4 Flash, which was retired on 2026-09-10.
+deepseek-flash, and the retired names that now route to it,
+have not been re-measured: a single probe on 2026-09-18 read +26 at
+high, not +79. Treat the column as history until the table
+is re-run.
Two of those levels are in no DeepSeek documentation at all.
none is documented only for the Responses API, but the chat
@@ -1392,19 +1402,21 @@ def jstr(s):
slug="pricing/",
crumb="pricing",
title="DeepSeek API pricing: the schedule, the peak hours, and the period right now",
- description="DeepSeek has billed peak/off-peak since 16:00 UTC on 2026-08-16: peak hours 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, at twice the off-peak rate, with weekends off-peak all day since 2026-08-22. The full schedule, the numbers per model in both periods, the flat card it replaced, and a strip that reads your clock and names the billing period you are in right now.",
- keywords="deepseek pricing, deepseek api pricing, deepseek price increase 2026, deepseek repricing, deepseek peak hours, deepseek off-peak pricing, deepseek peak off-peak billing, deepseek api cost, deepseek v4 flash price, deepseek v4 pro price, deepseek pricing 2026-08-16, deepseek new rate card, deepseek token price, deepseek weekend off-peak, deepseek peak weekdays only, deepseek v4 flash vision exp price",
+ description="DeepSeek has billed peak/off-peak since 16:00 UTC on 2026-08-16: peak hours 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, at twice the off-peak rate, with weekends off-peak all day since 2026-08-22. Since the V4.1 Flash release of 2026-09-10, deepseek-flash costs $0.003 / $0.15 / $0.60 per 1M off-peak. The full schedule, the numbers per model in both periods, the cards they replaced, and a strip that reads your clock and names the billing period you are in right now.",
+ keywords="deepseek pricing, deepseek api pricing, deepseek price increase 2026, deepseek repricing, deepseek peak hours, deepseek off-peak pricing, deepseek peak off-peak billing, deepseek api cost, deepseek v4 flash price, deepseek v4 pro price, deepseek pricing 2026-08-16, deepseek new rate card, deepseek token price, deepseek weekend off-peak, deepseek peak weekdays only, deepseek v4 flash vision exp price, deepseek flash price, deepseek v4.1 flash price, deepseek price cut 2026-09-10",
jsonld=faq([
("What does the DeepSeek API cost right now?",
- "It depends on the hour and the day. Per 1M tokens (cache hit / cache miss / output): deepseek-v4-flash is $0.007 / $0.22 / $0.66 off-peak and $0.014 / $0.44 / $1.32 peak; deepseek-v4-pro is $0.022 / $0.66 / $1.98 off-peak and $0.044 / $1.32 / $3.96 peak. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday; every other hour, and the whole weekend, is off-peak at half the peak rate. deepseek-v4-flash-vision-exp bills at exactly the deepseek-v4-flash rates."),
+ "It depends on the hour and the day. Per 1M tokens (cache hit / cache miss / output): deepseek-flash is $0.003 / $0.15 / $0.60 off-peak and $0.006 / $0.30 / $1.20 peak; deepseek-v4-pro is $0.022 / $0.66 / $1.98 off-peak and $0.044 / $1.32 / $3.96 peak. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday; every other hour, and the whole weekend, is off-peak at half the peak rate. The retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp are served by V4.1 Flash and bill at the deepseek-flash rates."),
("What are DeepSeek's peak hours?",
"01:00-04:00 and 06:00-10:00 UTC, Monday to Friday. The boundaries are defined in UTC; in Beijing time (UTC+8) they read 09:00-12:00 and 14:00-18:00. Every other hour is off-peak, at half the peak rate. Since 16:00 UTC on 2026-08-22 weekends bill off-peak all day, on the Beijing calendar, so the weekend starts at 16:00 UTC on Friday and peak is 35 hours a week rather than 49."),
("When did DeepSeek's peak/off-peak pricing start?",
"At 16:00 UTC on August 16, 2026, which is midnight in Beijing. Before that instant every hour billed at the flat card published 2026-08-02; from it, DeepSeek bills peak/off-peak on a new, higher card."),
("How much did DeepSeek raise its API prices?",
"Against the flat card it replaced, deepseek-v4-pro peak is 3x on cache-miss input ($0.435 to $1.32 per 1M), 4.6x on output ($0.87 to $3.96) and 12x on cache-hit input ($0.003625 to $0.044). Off-peak is half of each of those, so the smallest increase at any hour is still 1.5x on cache-miss input. The rise is steepest where the old card was cheapest."),
+ ("Did DeepSeek cut prices with V4.1 Flash?",
+ "Yes, on Flash only. From the V4.1 Flash release of 2026-09-10, deepseek-flash costs $0.003 / $0.15 / $0.60 per 1M tokens off-peak (cache hit / cache miss / output), down from $0.007 / $0.22 / $0.66 for deepseek-v4-flash: 57% less on cached input, 32% on uncached input, 9% on output. Peak is still double. deepseek-v4-pro did not change. DeepSeek published the date but no time; the deepseek CLI switches cards at 11:00 UTC that day, an inferred instant."),
("Is DeepSeek's off-peak price the same as the old price?",
- "No. Off-peak is half of peak, but on a new, higher base card: a flash cache-miss input token went from $0.14 to $0.22 per 1M off-peak and $0.44 peak. Even the cheapest hour now costs more than any hour did before the switch."),
+ "No. Off-peak is half of peak, but on a higher base than the flat card of 2026-08-02. Even after the V4.1 Flash cut, a flash cache-miss input token costs $0.15 per 1M off-peak against the flat card's $0.14, and output $0.60 against $0.28: the cheapest hour still costs more than any hour did before the switch."),
("How do I check the current DeepSeek billing period from the terminal?",
"Run `deepseek pricing`. It prints the period in effect (flat, off-peak or peak) with your local, UTC and Beijing time, when the period next changes, and both rate cards, all computed locally from the same schedule the CLI's cost estimates use. `deepseek pricing --json` emits the same as JSON."),
]),
@@ -1412,8 +1424,8 @@ def jstr(s):
Pricing
Since 16:00 UTC on 2026-08-16, what a DeepSeek token costs
depends on the hour you spend it in – and, since 2026-08-22, on the
-day of the week too. This page carries the schedule, both
-rate cards, and a strip that reads your clock – the same data the
+day of the week too. Since 2026-09-10 Flash costs less at every hour. This
+page carries the schedule, every card, and a strip that reads your clock – the same data the
CLI's estimates switch on, so the page and ds pricing can
never disagree.
@@ -1453,33 +1465,53 @@ def jstr(s):
hold it for Saturday instead.
The card
-USD per 1M tokens, in force since 2026-08-16 16:00 UTC. Off-peak is half
-of peak by construction, but on a higher base than the flat card it
-replaced – even the cheapest hour now costs more than any hour did
-before. A flash cache-miss input token went from $0.14 to $0.22 per 1M
-off-peak, and to $0.44 in peak hours:
+USD per 1M tokens, in force since the V4.1 Flash release of 2026-09-10.
+Off-peak is half of peak by construction. Flash got cheaper on every item
+– 57% on cached input, 32% on uncached input, 9% on output –
+and pro did not change:
Model Period Input (cached) Input (miss) Output deepseek-v4-flashoff-peak $0.007 $0.22 $0.66 peak $0.014 $0.44 $1.32 deepseek-flashoff-peak $0.003 $0.15 $0.60 peak $0.006 $0.30 $1.20 deepseek-v4-prooff-peak $0.022 $0.66 $1.98 peak $0.044 $1.32 $3.96
+The retired names deepseek-v4-flash and
+deepseek-v4-flash-vision-exp still work: V4.1 Flash serves them,
+at the deepseek-flash price.
+DeepSeek dated the cut and published no time, so the instant the
+CLI switches cards is inferred: 11:00 UTC on 2026-09-10. Our docs
+mirror fetched the old card at 04:50 UTC that day and the new one at 11:27
+UTC; 11:00 is the last whole hour between the two, so an estimate for a call
+made in that gap can overstate what it cost and never understate it.
The context cache stays the biggest
-lever: a cached input token costs about 1/30th of a miss, against 1/50th
-(flash) and 1/120th (pro) on the card this replaced. Prompt structure
-still dominates a bill – it is worth up to 30×, where the hour
-of the day is worth 2×.
+lever: a cached input token costs 1/50th of a miss on flash and 1/30th on
+pro. Prompt structure still dominates a bill – it is worth up to
+50×, where the hour of the day is worth 2×.
What it replaced
-The flat card published 2026-08-02, billed at every hour until the
-switch. It is here because the ledger
-keeps token counts rather than dollars, so a call made before 2026-08-16
-16:00 UTC still reprices under the card it was actually billed at:
+Two superseded cards. They are here because the
+ledger keeps token counts rather than
+dollars, so every call reprices under the card it was actually billed
+at.
+From 2026-08-16 16:00 UTC to the V4.1 cut, flash was
+deepseek-v4-flash on a card more than twice as dear on cached input; pro's
+rows were the same as today's:
+
+
+Model Period Input (cached) Input (miss) Output deepseek-v4-flashoff-peak $0.007 $0.22 $0.66 peak $0.014 $0.44 $1.32
+
+Before that, the flat card published 2026-08-02 billed at every hour
+until 2026-08-16 16:00 UTC:
Model Input (cached) Input (miss) Output
Peak output on pro is 4.6× that card; peak cached input is
-12× it. The published increase is steepest exactly where the old card
-was cheapest.
+12× it. The published increase was steepest exactly where the old
+card was cheapest, and the V4.1 cut has not brought flash back under it
+at any hour.
The same answer in the terminal
ds pricing # period right now, local + UTC + Beijing, both cards
@@ -1499,7 +1532,7 @@ def jstr(s):
Computed locally from the same schedule – no network call, nothing
spent. The cost estimates price each call with
the card in force at the moment it was made, which is what lets one ledger
-span both eras. The ledger stores exact
+span every era. The ledger stores exact
token counts rather than dollar amounts, so historical calls can always be
repriced under whatever card was, or becomes, real.
@@ -1513,7 +1546,8 @@ def jstr(s):
They are also checked against the bill. A 188,542-token cache-miss call
to pro, made off-peak on 2026-08-17, settled at 0.84 CNY
– 4.46 CNY per 1M against the published off-peak 4.5, where the old
-flat card would have made it 3.0. The page and the billing system agree.
+flat card would have made it 3.0. The page and the billing system agree.
+The V4.1 Flash card has not been checked against a bill yet.
""",
@@ -1637,16 +1671,17 @@ def jstr(s):
no reason to be chosen. The lever is price, and the gap is not small:
-Per 1M tokens v4-flash v4-pro GPT-5.6 Sol pro is cheaper by Per 1M tokens flash v4-pro GPT-5.6 Sol pro is cheaper by input, cache miss $0.22 / $0.44 $0.66 / $1.32 $5.00 7.6x / 3.8x input, cache hit $0.007 / $0.014 $0.022 / $0.044 $0.50 23x / 11x output $0.66 / $1.32 $1.98 / $3.96 $30.00 15x / 7.6x input, cache miss $0.15 / $0.30 $0.66 / $1.32 $5.00 7.6x / 3.8x input, cache hit $0.003 / $0.006 $0.022 / $0.044 $0.50 23x / 11x output $0.60 / $1.20 $1.98 / $3.96 $30.00 15x / 7.6x
-DeepSeek cells read off-peak / peak, on the
-card in force since 2026-08-16 16:00 UTC; they are a conversion of the RMB
+
DeepSeek cells read off-peak / peak: pro on the
+card in force since 2026-08-16 16:00 UTC, flash (deepseek-flash,
+V4.1) on the lower card of 2026-09-10. They are a conversion of the RMB
card (pro: ¥4.5 / ¥0.15 / ¥13.5 per 1M off-peak, double at peak)
at one consistent rate. GPT-5.6 Sol prices are from OpenAI's own listing.
The pricing page has the full schedule.
@@ -1713,10 +1748,12 @@ def jstr(s):
PAGES.append(dict(
slug="news/",
crumb="news",
- title="DeepSeek API news: the repricing is live, dsh ships, V4-Pro GA",
- description="What is changing around the DeepSeek API: peak/off-peak billing went live at 2026-08-16 16:00 UTC and is confirmed against a real bill, DeepSeek ships dsh (DeepSeek Harness), its official open-source agent harness, V4-Pro's official release (DeepSeek-V4-Pro-0813), V4-Flash's official release, and what each one means for a call.",
- keywords="deepseek harness, dsh, @deepseek-ai/dsh, install dsh, dsh plugins, dsh skills, dsh-plugin, dsh vs claude code, deepseek cli vs dsh, deepseek v4 pro release, deepseek v4 pro ga, deepseek-v4-pro-0813, deepseek api price increase, deepseek price rise 2026, deepseek peak hour pricing, deepseek peak off-peak billing, deepseek repricing 2026-08-16, deepseek new prices live, deepseek api news, deepseek api changelog, deepseek v4 flash release, deepseek pricing change",
+ title="DeepSeek API news: V4.1 Flash cuts the Flash price, the repricing is live, dsh ships",
+ description="What is changing around the DeepSeek API: DeepSeek-V4.1-Flash shipped 2026-09-10 as deepseek-flash, on a lower Flash card, with the V4 Flash names retired; peak/off-peak billing went live at 2026-08-16 16:00 UTC and is confirmed against a real bill, DeepSeek ships dsh (DeepSeek Harness), its official open-source agent harness, V4-Pro's official release (DeepSeek-V4-Pro-0813), V4-Flash's official release, and what each one means for a call.",
+ keywords="deepseek v4.1 flash, deepseek-flash, deepseek v4.1 flash price, deepseek price cut, deepseek-v4-flash retired, deepseek harness, dsh, @deepseek-ai/dsh, install dsh, dsh plugins, dsh skills, dsh-plugin, dsh vs claude code, deepseek cli vs dsh, deepseek v4 pro release, deepseek v4 pro ga, deepseek-v4-pro-0813, deepseek api price increase, deepseek price rise 2026, deepseek peak hour pricing, deepseek peak off-peak billing, deepseek repricing 2026-08-16, deepseek new prices live, deepseek api news, deepseek api changelog, deepseek v4 flash release, deepseek pricing change",
jsonld=faq([
+ ("What is DeepSeek-V4.1-Flash?",
+ "DeepSeek's Flash model since 2026-09-10, with native image input, called as deepseek-flash. It retired V4 Flash and V4 Flash Vision Exp; their names, deepseek-v4-flash and deepseek-v4-flash-vision-exp, are still accepted, are served by V4.1 Flash and bill at the Flash price. The Flash card fell to $0.003 / $0.15 / $0.60 per 1M tokens off-peak (cache hit / cache miss / output), from $0.007 / $0.22 / $0.66. deepseek-v4-pro continues with its card unchanged."),
("Is DeepSeek V4-Pro officially released?",
"Yes. On 2026-08-12 the model version on DeepSeek's Models & Pricing page changed to DeepSeek-V4-Pro-0813, ending the preview that had run since 2026-04-24. The model ID is unchanged (deepseek-v4-pro), the rate card is unchanged, and the release focuses on agentic post-training rather than new pretraining."),
("What is dsh (DeepSeek Harness)?",
@@ -1726,7 +1763,7 @@ def jstr(s):
("Did DeepSeek raise its API prices?",
"Yes. The increase took effect at 16:00 UTC on 2026-08-16 and is live: DeepSeek now bills peak/off-peak, with peak hours 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, at twice the off-peak rate; every other hour, and the whole weekend since 2026-08-22, is off-peak. The flat card is gone from the official pricing page. Against it, deepseek-v4-pro peak is 3x on cache-miss input, 4.6x on output and 12x on cache-hit input; off-peak is half of each. This resolved both the undated broad price rise announced on 2026-08-06 and the undated peak-hour policy announced in June 2026."),
("What are DeepSeek's current API prices?",
- "Per 1M tokens (cache hit / cache miss / output), on the card in force since 16:00 UTC on 2026-08-16: deepseek-v4-flash is $0.007 / $0.22 / $0.66 off-peak and $0.014 / $0.44 / $1.32 peak; deepseek-v4-pro is $0.022 / $0.66 / $1.98 off-peak and $0.044 / $1.32 / $3.96 peak. In RMB, pro is 0.15 / 4.5 / 13.5 yuan off-peak and 0.3 / 9 / 27 yuan peak. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday; every other hour, and the whole weekend, is off-peak."),
+ "Per 1M tokens (cache hit / cache miss / output), deepseek-flash is $0.003 / $0.15 / $0.60 off-peak and $0.006 / $0.30 / $1.20 peak, on the card in force since the V4.1 Flash release of 2026-09-10; deepseek-v4-pro, unchanged since 16:00 UTC on 2026-08-16, is $0.022 / $0.66 / $1.98 off-peak and $0.044 / $1.32 / $3.96 peak. In RMB, flash is 0.02 / 1 / 4 yuan off-peak and 0.04 / 2 / 8 yuan peak, and pro is 0.15 / 4.5 / 13.5 yuan off-peak and 0.3 / 9 / 27 yuan peak. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday; every other hour, and the whole weekend, is off-peak."),
("Which hours are cheapest on the DeepSeek API?",
"Every hour outside 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, which bill at half the peak rate — and since 2026-08-22 the whole weekend, on the Beijing calendar, so the cheap window opens at 16:00 UTC on Friday. Those peak windows are 09:00-12:00 and 14:00-18:00 Beijing time, the Chinese working day, so the whole European and American working day is off-peak. Batch and agent work that can be scheduled should run off-peak; the saving is exactly 2x."),
("Where can I follow DeepSeek API changes?",
@@ -1739,6 +1776,38 @@ def jstr(s):
live API where that is possible; the in-terminal feed is
ds docs changelog.
+2026-09-10 · V4.1 Flash, on a cheaper cardprice cut
+DeepSeek-V4.1-Flash is on the API as deepseek-flash, with
+native image input. V4 Flash and V4 Flash Vision Exp are retired; their names,
+deepseek-v4-flash and deepseek-v4-flash-vision-exp,
+still work, are served by V4.1 Flash, and bill at the Flash price.
+GET /models now lists exactly two ids: deepseek-flash
+and deepseek-v4-pro.
+The Flash card came down on every item. Off-peak, per 1M tokens, with peak
+still exactly double:
+
+
+Per 1M tokens, off-peak V4 flash, until 2026-09-10 deepseek-flash, from 2026-09-10 change input, cache hit $0.007 (¥0.05) $0.003 (¥0.02) −57% input, cache miss $0.22 (¥1.5) $0.15 (¥1) −32% output $0.66 (¥4.5) $0.60 (¥4) −9%
+
+The cut is deepest on the cheapest token, the mirror image of August: a
+cache hit on flash is back to 1/50th of a miss, where the flat card had it.
+Pro's card did not move, and DeepSeek says it will "continue providing API
+services for DeepSeek V4 Pro after September 14, 2026, with the billing method
+remaining unchanged." The peak windows and the weekend rule are unchanged
+too.
+The changelog dates the release and not the price change. ds
+pricing and the cost estimates switch cards at 11:00 UTC on
+2026-09-10, an inferred instant: our docs mirror fetched the old card
+at 04:50 UTC that day and the new one at 11:27 UTC, and a call estimated in
+that gap can only be overstated. The
+pricing page has both cards.
+
2026-08-22 · weekends are off-peak, all daynot in the changelog
From 16:00 UTC on 2026-08-22 – 00:00 Beijing on
Sunday 23 August – a Saturday or Sunday bills at the off-peak card for
@@ -1782,8 +1851,8 @@ def jstr(s):
ability, in their framing, close to Opus-4.8. Those are the vendor's numbers,
not ours; we have not run them.
Because it shipped after the 2026-08-16 switchover it has no flat
-card, so there is nothing to reprice for it before that date. It is in
-ds pricing from v0.5.1 on.
+card, so there is nothing to reprice for it before that date. It was retired
+on 2026-09-10; the name still works and bills as deepseek-flash.
2026-08-16 · the repricing is liveconfirmed against a bill
It landed on schedule. At 16:00 UTC on 2026-08-16 –
@@ -2316,7 +2385,7 @@ def jstr(s):
person draining it for everyone, enrolling asks your browser to solve a small
proof-of-work puzzle. That puzzle is the whole signup.
- - model deepseek-v4-flash
+ - model deepseek-flash (V4.1 Flash)
- per day 30 requests · 60k input · 20k output tokens
- privacy prompts are relayed to DeepSeek and are not stored or
logged by the gateway; only token counts and cost are recorded
diff --git a/site/commands/index.html b/site/commands/index.html
index f374744..73c18a8 100644
--- a/site/commands/index.html
+++ b/site/commands/index.html
@@ -105,7 +105,7 @@ chat
Flag Effect -m, --modeldeepseek-v4-flash (default) or deepseek-v4-pro-m, --modeldeepseek-flash (default) or deepseek-v4-pro-s, --systemSystem prompt, inline or @file --think on|offThinking mode. Default is the API's own, which is on -e, --effortlow, high or maxWhat DeepSeek actually costs: cache math and a local usage ledger
-
-
+
+
@@ -13,7 +13,7 @@
-
+
@@ -21,7 +21,7 @@
-
+
@@ -87,7 +87,7 @@
Cost
DeepSeek's headline feature is a disk-backed context cache that
-makes a repeated prompt prefix roughly thirty times cheaper. Since
+makes a repeated prompt prefix thirty to fifty times cheaper. Since
2026-08-16 a second lever sits on top of it: the hour of the day, worth 2×.
Neither is visible unless something is counting – so this counts.
@@ -101,22 +101,23 @@ The rate card
Model Period Input (cached) Input (miss) Output deepseek-v4-flashoff-peak $0.007 $0.22 $0.66 peak $0.014 $0.44 $1.32 deepseek-v4-flash-vision-expoff-peak $0.007 $0.22 $0.66 peak $0.014 $0.44 $1.32 deepseek-flashoff-peak $0.003 $0.15 $0.60 peak $0.006 $0.30 $1.20 deepseek-v4-prooff-peak $0.022 $0.66 $1.98 peak $0.044 $1.32 $3.96
-ds models prints the card in force next to the live model
+
The retired names deepseek-v4-flash and
+deepseek-v4-flash-vision-exp still work: V4.1 Flash serves them,
+at the deepseek-flash price. ds models prints the card in force next to the live model
list, so the price is on screen when you pick, and ds pricing
prints the full schedule with the period you are in right now.
What the cache is worth
-On flash, a cache hit costs 1/31st of a miss per token.
-The same 2,900-token prompt, sent twice:
+On flash, a cache hit costs 1/50th of a miss per token;
+on pro, 1/30th. The same 2,900-token prompt, sent twice, on the V4 flash
+card of the day, where the ratio was 1/31st:
$ ds chat "..." --system @prefix.txt
@@ -131,7 +132,8 @@ What the cache is worth
same prefix again. It is 15× and not 31× because the 3% that
changed – the tail of the prompt, and the question itself –
still bills at the miss rate: the realised saving is always the rate ratio
-times how much of the prompt you kept identical. Hence the practical rule:
+times how much of the prompt you kept identical. On today's flash card the
+same call drops about 20×. Hence the practical rule:
put the stable part of a prompt first – same system
prompt, same files, in the same order – and let the variable part come
last.
@@ -157,7 +159,7 @@ The thinking surcharge
---effortflash pro thinking --effortV4 flash pro thinking none+0 +0 off entirely minimal, low+0 +0 on The thinking surcharge
+The flash column is V4 Flash, which was retired on 2026-09-10.
+deepseek-flash, and the retired names that now route to it,
+have not been re-measured: a single probe on 2026-09-18 read +26 at
+high, not +79. Treat the column as history until the table
+is re-run.
Two of those levels are in no DeepSeek documentation at all.
none is documented only for the Responses API, but the chat
diff --git a/site/formats/index.html b/site/formats/index.html
index 38f3b06..95e6731 100644
--- a/site/formats/index.html
+++ b/site/formats/index.html
@@ -30,7 +30,7 @@