Pricing estimates#
Decant estimates token costs at ingest using published first-party API rates.
Gemini 4 Neon rates were verified on November 18, 2026, against Google's
standard Gemini Developer API pricing.
GPT-6.2 Astra and Sol rates were verified on November 17, 2026, against
OpenAI's pricing page, model pages, and launch announcement below.
On November 16, 2026, every Anthropic, OpenAI, Gemini, Z.ai, Meta, DeepSeek,
MiniMax, MiMo, Qwen, StepFun, Tencent, and Reflection rate then in the table
was checked against the provider pages listed here and found unchanged, and the
new Gemini 4 Carbon and Hydrogen, GPT-6 Cyber, GPT-5.5 Cyber Preview,
Qwen4-Max, Muse Ember, MiniMax-M3.1-Flash, Hy4, Step 3.7 Flash, and xAI Grok
rates were added. The dates that follow record when each rate was first
verified. Claude Fable 5.2 and Mythos 5.2 rates, including their fast-mode
rates, were verified on November 9, 2026. Claude Haiku 5.5 rates, the Claude Sonnet 5.5 cache-read rate, and fast-mode
rates were verified on October 7, 2026, and all other Anthropic rates on
September 28, 2026, against the Anthropic pricing and model deprecation pages
below. GPT-6.1 Terra rates were verified on October 26, 2026, GPT-6.1 Sol
rates on September 29, 2026, GPT-6 Sol and Luna rates on September 22, 2026,
and other current OpenAI coding-agent rates retain their September 9, 2026
verification. Legacy OpenAI rates were checked September 2, 2026, Gemini 4
Argon rates November 5, 2026, and other Gemini rates September 28, 2026.
Z.ai GLM, Meta Muse, DeepSeek, Moonshot Kimi, MiniMax, Xiaomi MiMo, Alibaba
Qwen, StepFun, Tencent Hy, and Reflection Beam rates were verified on October
26, 2026. The open-weight models in that list were chosen from the
Artificial Analysis open-weight leaderboard,
which Decant uses only to decide which models to price, never as the source of
a rate. Kimi K3.1 Pro rates were verified on October 25, 2026.
- Anthropic model and prompt-cache pricing
- Claude Fable 5.2 model details
- Claude Opus 5.5 model details
- Claude Sonnet 5.5 model details
- Claude Haiku 5.5 model details
- Introducing Claude Haiku 5.5 (Haiku 5.5 rates and the Sonnet 5.5 cache-read reduction)
- Anthropic fast mode pricing
- Anthropic model deprecations
- OpenAI API pricing
- OpenAI model pages
- OpenAI GPT-6.2 Astra pricing
- OpenAI GPT-6.2 Sol pricing
- Introducing GPT-6.2 Astra and Sol
- OpenAI GPT-6 Astra pricing
- OpenAI GPT-6 Sol pricing
- OpenAI GPT-6.1 Sol pricing
- Introducing GPT-6.1 Sol
- OpenAI GPT-6.1 Terra pricing
- OpenAI GPT-6 Luna pricing
- OpenAI GPT-6 Cyber pricing (checked November 16, 2026)
- OpenAI GPT-5.6 Sol pricing (promotion end date, checked November 16, 2026)
- OpenAI Codex credit pricing
- OpenAI model deprecations
- OpenAI GPT-3.5 Turbo launch pricing
- OpenAI GPT-3.5 Turbo June 2023 price update
- Gemini Developer API pricing (checked November 16, 2026)
- Gemini 4 Neon pricing (checked November 18, 2026)
- Gemini 4 Neon reasoning-level model IDs
- Gemini models (Gemini 4 Argon, Carbon, and Hydrogen model codes, checked November 16, 2026)
- Gemini API changelog (Gemini 3.5 Flash and 3.7 Flash deprecations, checked November 16, 2026)
- Z.ai pricing (checked November 16, 2026)
- Meta Model API pricing (checked November 16, 2026)
- DeepSeek API pricing (checked November 16, 2026)
- Kimi API pricing and context caching (checked October 26, 2026)
- Kimi K3.1 Pro API pricing (checked October 25, 2026)
- MiniMax pricing (checked November 16, 2026)
- Xiaomi MiMo pay-as-you-go pricing (checked November 16, 2026)
- Alibaba Cloud Model Studio pricing and context cache (checked November 16, 2026)
- QwenCloud Qwen3.8 2.4T A95B model page (its cached-input rate, checked November 16, 2026)
- QwenCloud Qwen4-Max model page and Model Studio new models (checked November 16, 2026)
- StepFun pricing (checked November 16, 2026)
- Tencent Cloud TokenHub model pricing (checked November 16, 2026)
- Reflection beta pricing and models (checked November 16, 2026)
- xAI pricing and May 15 retirement (checked November 16, 2026)
All dollar amounts below are USD per million tokens. A dash means that the
provider does not publish a first-party API token price for that model slug.
Coding-agent model rates#
| Provider models | Input | Cached input | Cache write | Output |
|---|---|---|---|---|
| Claude Fable 5.2, Mythos 5.2 | $8.00 | $0.20 | $10.00 / $16.00 | $40.00 |
| Claude Fable 5.1, Mythos 5.1 | $10.00 | $0.25 | $12.50 / $20.00 | $50.00 |
| Claude Fable 5, Mythos 5 | $10.00 | $1.00 | $12.50 / $20.00 | $50.00 |
| Claude Opus 5.5 | $4.00 | $0.20 | $5.00 / $8.00 | $20.00 |
| Claude Opus 5, 4.8, 4.7, 4.6, 4.5 | $5.00 | $0.50 | $6.25 / $10.00 | $25.00 |
| Claude Opus 4.1, 4 | $15.00 | $1.50 | $18.75 / $30.00 | $75.00 |
| Claude Sonnet 5.5 | $2.00 | $0.10 | $2.50 / $4.00 | $10.00 |
| Claude Sonnet 5 | $2.00 | $0.20 | $2.50 / $4.00 | $10.00 |
| Claude Sonnet 4.6, 4.5, 4 | $3.00 | $0.30 | $3.75 / $6.00 | $15.00 |
| Claude Haiku 5.5 | $0.10 | $0.01 | $0.125 / $0.20 | $0.50 |
| Claude Haiku 4.5 | $1.00 | $0.10 | $1.25 / $2.00 | $5.00 |
| Claude Haiku 3.5 | $0.80 | $0.08 | $1.00 / $1.60 | $4.00 |
| Claude Opus 3, Sonnet 3.7, 3.5, 3, Haiku 3 | โ | โ | โ | โ |
| GPT-6 Cyber | $20.00 | $2.00 | $25.00 | $100.00 |
| GPT-6.2 Astra | $8.00 | $0.40 | $10.00 | $40.00 |
| GPT-6.2 Sol | $2.00 | $0.10 | $2.50 | $10.00 |
| GPT-6 Astra | $10.00 | $1.00 | $12.50 | $50.00 |
| GPT-6 Sol | $2.00 | $0.20 | $2.50 | $10.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $2.50 | $10.00 |
| GPT-6.1 Terra | $1.00 | $0.05 | $1.25 | $5.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.125 | $0.50 |
| GPT-5.6 Sol, GPT-5.6, Daybreak Blue | $4.00 | $0.40 | $5.00 | $20.00 |
| GPT-5.6 Terra | $2.00 | $0.20 | $2.50 | $12.00 |
| GPT-5.6 Luna | $0.20 | $0.02 | $0.25 | $1.20 |
| GPT-5.6 Cyber, Daybreak Red | $12.50 | $1.25 | $15.625 | $75.00 |
| GPT-5.5 Cyber Preview | $12.50 | $1.25 | no additional fee | $75.00 |
| GPT-5.5 | $5.00 | $0.50 | no additional fee | $30.00 |
| GPT-5.4 | $2.50 | $0.25 | no additional fee | $15.00 |
| GPT-5.4 mini | $0.75 | $0.075 | no additional fee | $4.50 |
| GPT-5.3-Codex, GPT-5.2-Codex | $1.75 | $0.175 | no additional fee | $14.00 |
| GPT-5.1-Codex, GPT-5.1-Codex-Max, GPT-5-Codex | $1.25 | $0.125 | no additional fee | $10.00 |
| GPT-5.1-Codex-mini | $0.25 | $0.025 | no additional fee | $2.00 |
| codex-mini-latest | $1.50 | $0.375 | no additional fee | $6.00 |
| codex-auto-review, GPT-5.3-Codex-Spark, GPT-5-Codex-mini, GPT-5.4-cyber | โ | โ | โ | โ |
| Gemini 4 Argon, Gemini 4 Carbon | $4.00 | $0.20 | $4.00-equiv | $20.00 |
| Gemini 4 Neon | $3.00 | $0.15 | $3.00-equiv | $15.00 |
| Gemini 4 Hydrogen | $1.50 | $0.075 | $1.50-equiv | $7.50 |
| Gemini 3.5 Flash | $1.50 | $0.15 | $1.50-equiv | $9.00 |
| Gemini 3.5 Flash-Lite | $0.30 | $0.03 | $0.30-equiv | $2.50 |
| Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash | $0.75 | $0.075 | $0.75-equiv | $3.75 |
| Gemini 3 Flash Preview | $0.50 | $0.05 | $0.50-equiv | $3.00 |
| Gemini 3.1 Pro Preview | $2.00 | $0.20 | $2.00-equiv | $12.00 |
| Gemini 3.1 Flash-Lite | $0.25 | $0.025 | $0.25-equiv | $1.50 |
| Gemini 2.5 Pro | $1.25 | $0.125 | $1.25-equiv | $10.00 |
| Gemini 2.5 Flash | $0.30 | $0.03 | $0.30-equiv | $2.50 |
| Gemini 2.5 Flash-Lite | $0.10 | $0.01 | $0.10-equiv | $0.40 |
| Gemini image, TTS, live, transcribe, native-audio, computer-use variants | โ | โ | โ | โ |
| GLM-5.3, GLM-5.2, GLM-5.1 | $1.40 | $0.26 | no additional fee | $4.40 |
| GLM-5.3-FlashX, GLM-5.3-FlashX-Legal | $0.37 | $0.075 | no additional fee | $1.25 |
| GLM-5.3-Flash | $0.15 | $0.03 | no additional fee | $0.50 |
| GLM-5 | $1.00 | $0.20 | no additional fee | $3.20 |
| GLM-4.7, GLM-4.6, GLM-4.5 | $0.60 | $0.11 | no additional fee | $2.20 |
| GLM-4.5-X | $2.20 | $0.45 | no additional fee | $8.90 |
| GLM-4.5-AirX | $1.10 | $0.22 | no additional fee | $4.50 |
| GLM-4.5-Air | $0.20 | $0.03 | no additional fee | $1.10 |
| GLM-4.7-FlashX | $0.07 | $0.01 | no additional fee | $0.40 |
| GLM-4-32B-0414-128K | $0.10 | $0.10 | no additional fee | $0.10 |
| GLM-4.7-Flash, GLM-4.5-Flash | free | free | free | free |
| GLM-5.3-HighSpeed, GLM-5.2-HighSpeed, GLM-5-Turbo, GLM-5V-Turbo | โ | โ | โ | โ |
| Muse Wonder 1.0 | $4.00 | $0.20 | no additional fee | $20.00 |
| Muse Spark 1.3, 1.2, 1.1 | $1.25 | $0.15 | no additional fee | $4.25 |
| Muse Spark 1.3 Contributor, 1.2 Contributor | $0.10 | $0.002 | no additional fee | $0.20 |
| Muse Ember 1.0 | $0.10 | $0.004 | no additional fee | $0.50 |
| Muse Wonder Full, Muse Glimmer, Muse Image, Muse Voice | โ | โ | โ | โ |
DeepSeek Flash (deepseek-flash, deepseek-v4-flash) | $0.30 | $0.006 | no additional fee | $1.20 |
| DeepSeek V4 Pro | $1.32 | $0.044 | no additional fee | $3.96 |
| Kimi K2.7 Code, Kimi K2.6 | $0.95 | $0.19 / $0.16 | no additional fee | $4.00 |
| Kimi K2.7 Code HighSpeed | $1.90 | $0.38 | no additional fee | $8.00 |
| Kimi K3.1 Pro | $2.00 | $0.20 | $2.00 / $4.00 | $10.00 |
| MiniMax-M3 | $0.30 | $0.06 | no additional fee | $1.20 |
| MiniMax-M3.1-Flash | $0.15 | $0.015 | no additional fee | $0.60 |
| MiMo-V2.6-Pro | $0.435 | $0.0036 | free | $0.87 |
| MiMo-V2.6-Flash | $0.14 | $0.0028 | free | $0.28 |
| MiMo-V2.6-Pro-UltraSpeed | $4.35 | $0.036 | free | $8.70 |
| Qwen4-Max | $1.50 | $0.075 | $1.875 / $3.00 | $4.50 |
| Qwen3.8 2.4T A95B | $2.00 | $0.25 | $2.50 | $6.00 |
| Qwen3.8 27B | $0.50 | $0.10 | $0.625 | $3.00 |
| Qwen3.5 397B A17B | $0.60 | no cache | no cache | $3.60 |
| Qwen3.5 122B A10B | $0.40 | no cache | no cache | $3.20 |
| Qwen3.6 35B A3B | $0.375 | no cache | no cache | $2.25 |
| Qwen3.6 27B | $0.60 | no cache | no cache | $3.60 |
| Qwen3.5 35B A3B | $0.25 | no cache | no cache | $2.00 |
| Qwen3.5 27B | $0.30 | no cache | no cache | $2.40 |
| Step 5 Preview | $1.00 | $0.10 | no additional fee | $2.70 |
| Step 3.7 Flash | $0.20 | $0.04 | no additional fee | $1.15 |
| Step 3.5 Flash | $0.10 | $0.02 | no additional fee | $0.30 |
| Hy4, Hy4 preview | $0.834 | $0.042 | no additional fee | $2.501 |
| Hy4 flash | $0.07 | $0.003 | no additional fee | $0.281 |
| Hy3 | $0.132 | $0.033 | no additional fee | $0.528 |
| Grok 4.7, Grok 4.6 | $2.00 | $0.50 | no additional fee | $6.00 |
| Grok 4.5 | $2.00 | $0.30 | no additional fee | $6.00 |
Grok 4.3, Grok 4.20 (0309 reasoning, non-reasoning, multi-agent) | $1.25 | $0.20 | no additional fee | $2.50 |
| Grok Build 0.1 | $1.00 | $0.20 | no additional fee | $2.00 |
| Beam 1.0 | free | free | free | free |
deepseek-chat, deepseek-reasoner, Kimi K2.5, Kimi K3 and K3.1, MiniMax M2.x, MiMo V2.5, Qwen3.8-Flash, Qwen3.8-Flash-Next, qwen4-max-preview, hy3-preview, grok-code-fast-1 | โ | โ | โ | โ |
| Ling 3.1 Flash, Muse Spark Open, Nemotron 3.5 Super VL, Nemotron 3 Ultra, Inkling, K2 Horizon | โ | โ | โ | โ |
The two cache-write figures for Claude, Kimi K3.1 Pro, and Qwen4-Max are the
5-minute and 1-hour rates. Kimi K3.1 Pro uses flat rates across its full context window.
OpenAI cache writes before GPT-5.6 have no additional fee. The GPT-5.6, GPT-6,
GPT-6.1, and GPT-6.2 models above bill cache writes at 1.25 times the
uncached-input rate, which OpenAI lists as the 30-minute cache write. OpenAI
also lists a 12-hour cache write at $20.00 for GPT-6.2 Astra and $5.00 for
GPT-6.2 Sol. Codex does not report which retention a write used, so Decant
bills every write at the 30-minute rate. For Gemini models with explicit
caching, Google bills cache storage by the token-hour. Decant approximates
that charge at the input rate for typical session-length holds. Gemini 4 Neon
uses implicit caching. If an imported record reports Neon tokens separately as
cache writes, Decant values them at the input rate with no storage surcharge.
The low, medium, high, and xhigh Neon IDs use the same standard API
rates.
Z.ai and Meta charge no separate cache-write fee, so Decant bills cache writes
for GLM and Muse models at the input rate. Z.ai lists cached-input storage as
free for a limited time, so Decant adds no holding charge. Z.ai publishes
GLM-4.7-Flash and GLM-4.5-Flash as free, so Decant prices them at zero rather
than leaving them unpriced. The HighSpeed GLM models are only offered through
the GLM Coding Plan, and GLM-5-Turbo and GLM-5V-Turbo are absent from Z.ai's
pricing page, so they stay unpriced. Muse Wonder Full is only available
through a Muse Code subscription, and Muse Glimmer only through third-party
routers, so neither has a first-party API rate. OpenRouter's unversioned
meta/muse-wonder bills at Muse Wonder 1.0's rates, so Decant maps it there.
Router prefixes such as z-ai/ and meta/, hosted routing suffixes such as
OpenRouter's :free and Ollama's :cloud, and promotional -free copies are
priced at the model's list rate. Other tags, such as an Ollama or LM Studio
quantization like :q4_K_M, name a local build with no API spend, so Decant
leaves them unpriced. The same rules apply to the DeepSeek, Kimi, MiniMax,
MiMo, Qwen, StepFun, Hy, Grok, and Beam IDs, so moonshotai/kimi-k2.6,
x-ai/grok-4.7, and deepseek-v4-pro:cloud reach the list rate and
qwen3.8:27b stays unpriced. Meta lists Muse Ember 1.0's rate above for
prompts up to 200,000 tokens and three times it for longer prompts. Decant
applies the shorter-prompt rate only, so estimates for longer requests are low.
DeepSeek charges the rates above during peak hours, 01:00 to 04:00 and 06:00
to 10:00 UTC on weekdays, and half those rates at all other times. Decant
prices every DeepSeek request at the peak rate, so its estimates are an upper
bound. DeepSeek now serves deepseek-v4-flash and
deepseek-v4-flash-vision-exp as deepseek-flash at the Flash rate.
OpenRouter names the same models by the version DeepSeek serves, so
deepseek-v4.1-flash-1017 and deepseek-v4.1-flash bill at the Flash rate
and deepseek-v4-pro-0813 at the V4 Pro rate. deepseek-chat and
deepseek-reasoner were scheduled for discontinuation on July 24, 2026 and no
longer appear on the pricing page, so they stay unpriced. DeepSeek's
Anthropic-compatible endpoint maps Claude model names to DeepSeek models.
Decant prices any request recorded under a Claude ID at Claude rates, which
for those requests is far above what DeepSeek bills.
MiniMax describes the MiniMax-M3 rates above as a permanent 50% discount from
its list price, and Decant uses the discounted rates it actually charges.
Prompts above 512,000 tokens bill at twice those rates, and the Priority
service tier at 1.5 times. Decant applies neither modifier, so estimates for
those requests are low. MiniMax lists no cache-write fee for
MiniMax-M3.1-Flash. Moonshot's Kimi K2 models do not support cache writes,
and Moonshot retires all K2 models by the end of December 2026. Kimi K3 and
K3.1 are not open-weight and are not yet priced. Kimi K3.1 Pro has its own
published rates and is priced above.
Xiaomi lists MiMo cache writes as free for a limited time with no end date, so
Decant bills them at zero. Alibaba prices Qwen at the international
(Singapore) rates above. Implicit cache hits bill at the cached rate, and only
explicit caching creates a cache entry, at 125% of input. The Qwen3.5 and
Qwen3.6 open models do not support caching, so any cache reads reported for
them bill at the input rate. Qwen3.8-Flash-Next has no first-party API, and
the hosted qwen3.8-flash built on it is a commercial model, so neither is
priced. Qwen4-Max, released November 16, 2026, is the first Qwen flagship
without open weights. Decant prices it because it is Alibaba's current
flagship. Its implicit and explicit cache reads both bill at $0.075, and
explicit cache writes bill by TTL. Its preview, qwen4-max-preview, was
offered only through Alibaba's Token Plan with no per-token rate, so it stays
unpriced. Alibaba has said smaller open-weight Qwen4 models will follow but
has not named or priced them.
Tencent lists the Hy4 and Hy4 flash rates above as list prices and charges
half of them for usage through December 31, 2026 for Hy4 and through
November 30, 2026 for Hy4 flash, both on Beijing time. Decant prices them at
list rate, so estimates for usage during those promotions are twice what
Tencent bills. StepFun's cache-miss input price includes writing to the cache.
xAI bills every token in a request at twice the rates above once the prompt
reaches 200,000 tokens, and charges 1.1 times for its US regional endpoint.
Decant applies neither modifier. xAI retired grok-code-fast-1 on May 15,
2026 and now routes its requests to Grok Build 0.1 without saying what they
bill, so that ID and its aliases stay unpriced.
Reflection's Beam API is free during its beta, subject to daily limits, so Decant prices Beam-1.0-501B-A23B at zero. The preview ID
Beam-501B-A23B is not documented as an alias and stays unpriced.
Several open-weight models with high Artificial Analysis scores have no USD
per-token first-party rate. Ant Group publishes Ling 3.1 Flash prices only in
CNY, and Decant does not convert currencies. Meta does not serve Muse Spark
Open on its Model API, and it is a separate model from Muse Spark 1.3. NVIDIA
offers Nemotron only through free trial endpoints and per-GPU licenses. The
Institute of Foundation Models publishes no price for K2 Horizon.
Thinking Machines publishes two price lists for Inkling without saying which
one applies to its Anthropic-compatible endpoint. All of these stay unpriced.
Anthropic made Claude Sonnet 5's $2.00 input and $10.00 output rates standard
on September 1, 2026, so the previously announced increase did not take
effect. Claude Sonnet 5.5 launched on September 28, 2026 at the same rates as
Sonnet 5. Anthropic halved its cache-read rate to $0.10 starting October 7,
2026, per the Claude Haiku 5.5 announcement; the pricing page's model table
still listed $0.20 when checked that day. Claude Opus 5.5 has a lower input,
output, and cache-read rate than Opus 5. Claude Haiku 5.5 launched on October
7, 2026 with prompt-length pricing: the rates above apply to prompts up to
100,000 tokens, and every rate is five times higher for longer prompts. Claude Fable 5.1 and Mythos 5.1 keep
Fable 5's input, cache-write, and output rates but reduce cache reads from
$1.00 to $0.25. Claude Fable 5.2 and Mythos 5.2 launched on November 9, 2026
at rates 20% below Fable 5.1, with the same 0.025x cache-read multiplier.
OpenAI describes the
current GPT-5.6 Sol rate as promotional through November 30, 2026. GPT-6 Cyber
and GPT-5.5 Cyber Preview are available only through OpenAI's Daybreak trusted
access program. GPT-5.5 Cyber Preview lists no cache-write fee and has its own
rate, so Decant no longer bills it at GPT-5.5's rate.
Google prices Gemini 4 Argon and Gemini 4 Carbon at $2.00 input, $0.10 cached,
and $10.00 output through November 30, 2026. The rate above applies from
December 1, 2026, and Decant reprices stored Argon and Carbon usage at it on
the next sync, as it does for every rate change. Gemini 4 Hydrogen launched on
November 12, 2026 with no promotion. The Gemini 4 models support implicit
caching only.
Google's Gemini 3.8 Flash, 3.7 Flash, and 3.6 Flash listed above carry a promotional rate
through December 31, 2026; they rise to $1.50 input / $0.15 cached / $7.50
output on January 1, 2027. Google deprecated Gemini 3.7 Flash and Gemini 3.5
Flash on October 8, 2026 and routes their requests to Gemini 3.8 Flash and
3.6 Flash. It does not say what routed requests bill, so Decant keeps each
model's last published rate. Google no longer lists the Gemini 2.0 line or Gemini 3
Pro Preview on the pricing page, and Gemini versions without a published rate stay unpriced
rather than inheriting a neighbor's rate.
Claude 3 model IDs such as claude-3-5-haiku-20241022 place the tier after
the version, so Decant recognizes that form as well as tier-first IDs like
claude-haiku-4-5. Anthropic's pricing page no longer lists the fully retired
Claude Opus 3, Sonnet 3.7, Sonnet 3.5, Sonnet 3, or Haiku 3, so Decant leaves
them unpriced instead of assigning a current tier's rate. Haiku 3.5, Opus 4.1,
Opus 4, and Sonnet 4 are retired on the Claude API but remain listed because
partner clouds still serve them.
codex-auto-review is a hidden routing slug, not a public billable model ID.
OpenAI does not document which underlying model or rate applies. Decant leaves
it and other unpublished Codex slugs unpriceable instead of guessing a nearby
model's price.
For gpt-3.5-turbo-0301, Decant uses the final published GPT-3.5 Turbo rate
of $1.50 input and $2.00 output. OpenAI's current deprecations table displays
$15.00 and $20.00, but those figures conflict with the contemporaneous launch
and June 2023 pricing announcements linked above.
Request tiers#
Decant prices each API request at the model that served it, and applies two
request-level modifiers where the source records what they need:
- Fast mode. A Claude request whose usage reports
speed: "fast"is billed
at twice the standard rate for every token category, cache reads and writes
included, on Claude Fable 5.2 and Mythos 5.2 ($16.00 input / $80.00 output),
Opus 5.5 ($8.00 / $40.00), Opus 5, and Opus 4.8 ($10.00 / $50.00). Opus 4.6
bills fast requests at standard rates. - Claude Haiku 5.5 long prompts. A request whose prompt (uncached input
plus cache reads and cache writes) exceeds 100,000 tokens is billed at five
times every rate above.
Sources without per-request usage, such as Codex rollouts, are priced from
session totals at the session's model and standard rates. Copilot CLI sessions
are priced from per-model totals at standard rates. The totals are the summed
usage records when those reach every model's snapshot total, and the snapshots
otherwise. See token and cost totals.
Copilot Chat in VS Code records partial usage per turn, and each turn is priced
at the model that served it. Output sums every model call in the turn. Input is
only the last call's prompt, so a turn with tool calls leaves its earlier
prompts uncounted. VS Code writes no cache split, so cached prompt tokens are
charged at the full input rate. Builds from April and May 2026 record output
but no prompt size, so their turns are priced on output alone. Sessions from
earlier builds record no usage, and Decant shows their tokens and cost as
unavailable, not free. These costs are rough API list-price estimates, not
Copilot premium requests or credits. See
token and cost totals.
Scope and historical behavior#
The estimates use standard global API token rates. They do not attempt to
convert ChatGPT subscription usage or Codex credits to dollars, and they do not
apply Batch, Flex, OpenAI Fast or Ultrafast, regional-processing,
data-residency, or partner-cloud modifiers. Requests made through subscription
routes such as the GLM Coding Plan are estimated at the model's API list rate.
GitHub Copilot CLI costs likewise estimate the underlying model's API list price, not Copilot
subscription charges or premium requests.
OpenAI applies long-context rates above 272,000 prompt tokens for eligible API
models. Codex reports cumulative session usage rather than per-request usage,
so Decant cannot tell whether an individual request crossed that threshold.
These estimates use short-context rates and may be low for qualifying
requests. Claude 4.6 and later models other than Haiku 5.5 include their full
context window at the standard rate.
Costs are stored when transcripts are ingested and checked against current
pricing on every sync, including unchanged and archived sessions. Sync updates
stale session costs and persisted activity cost components together without
rereading transcripts or replacing user metadata. Later syncs leave matching
costs untouched.
OpenCode sessions are priced with Decant's first-party rates whenever the model
is supported, including the GLM, Muse, and other lab rates above. For a model
Decant cannot price, Decant falls back to the per-request cost OpenCode
recorded (the reported cost). A recorded cost of 0 leaves the model unpriced,
and a model that later gains a Decant rate switches back to Decant pricing.
Each session carries a cost_source label:
decant: priced entirely with Decant's rates.reported: priced only with costs OpenCode recorded.mixed: some models priced by Decant and others by reported cost.
Aggregate dollar totals therefore include Decant's estimates for supported
models, plus reported costs for OpenCode sessions with models Decant cannot
price. Reported costs are not first-party estimates, and token-economics totals
report them separately as reported_cost_usd.