No positions in the markets we measure AS OF 2026-08-22 BY 07:30 ET
CCIR Compute Credit
Index Research
Benchmark · USD per 1M tokens · 173 models as of 2026-08-22

Token Prices

posted price by model · input & output · USD per 1M tokens · 2026-08-22
input output
$0 $10 $20 $30 $40 $50 DeepSeek V4 Flash — input $0.44 per 1M tokens (first-party posted) DeepSeek V4 Flash — output $1.32 per 1M tokens (first-party posted) $0.44 $1.32 DeepSeek V4 Flash Qwen3 Coder 480B — input $1.60 per 1M tokens (first-party posted) Qwen3 Coder 480B — output $6.40 per 1M tokens (first-party posted) $1.60 $6.40 Qwen3 Coder 480B Moonshot Kimi K3 — input $3.00 per 1M tokens (first-party posted) Moonshot Kimi K3 — output $15 per 1M tokens (first-party posted) $3.00 $15 Moonshot Kimi K3 OpenAI GPT-5.6 Sol — input $4.00 per 1M tokens (first-party posted) OpenAI GPT-5.6 Sol — output $20 per 1M tokens (first-party posted) $4.00 $20 OpenAI GPT-5.6 Sol Anthropic Opus 5 — input $5.00 per 1M tokens (first-party posted) Anthropic Opus 5 — output $25 per 1M tokens (first-party posted) $5.00 $25 Anthropic Opus 5 Anthropic Fable 5 — input $10 per 1M tokens (first-party posted) Anthropic Fable 5 — output $50 per 1M tokens (first-party posted) $10 $50 Anthropic Fable 5

Rows carrying an n are served medians across that many providers. The rest are the model owners' posted prices. Two constructions, never pooled.

01

Frontier posted — first-party

from the owners' price lists

OpenAIstandard tier

ModelInputCachedOutput30dLast reprice
gpt-5.6-sol $4.00 $0.40 $20.00 2026-08-22 (-33%)
gpt-5.6-terra $2.00 $0.20 $12.00 2026-07-30 (-20%)
gpt-5.6-luna $0.20 $0.02 $1.20 2026-07-30 (-80%)
gpt-5.5 $5.00 $0.50 $30.00 none in record
gpt-5.5-pro $30.00 — $180.00 none in record
gpt-5.4-mini $0.75 $0.07 $4.50 none in record
gpt-5.4-nano $0.20 $0.02 $1.25 none in record
gpt-4o $2.50 $1.25 $10.00 none in record
gpt-4o-mini $0.15 $0.07 $0.60 none in record
o3-pro $20.00 — $80.00 none in record
o3 $2.00 $0.50 $8.00 none in record
o4-mini $1.10 $0.28 $4.40 none in record
gpt-4-turbo-2024-04-09 $10.00 — $30.00 none in record
gpt-3.5-turbo-instruct $1.50 — $2.00 none in record

Anthropicmodel table

ModelInputCachedOutput30dLast reprice
Claude Fable 5 $10.00 $1.00 $50.00 none in record
Claude Mythos 5 $10.00 $1.00 $50.00 none in record
Claude Opus 5 $5.00 $0.50 $25.00 none in record
Claude Sonnet 5 $2.00 $0.20 $10.00 none in record
Claude Haiku 4.5 $1.00 $0.10 $5.00 none in record

Googlestandard tier

ModelInputCachedOutput30dLast reprice
gemini-3.7-flash $0.75 $0.07 $3.75 none in record
gemini-3.5-flash-lite $0.30 $0.03 $2.50 none in record
gemini-3.1-pro-preview $2.00 $0.20 $12.00 none in record
gemini-3-flash-preview $0.50 $0.05 $3.00 none in record
gemini-2.5-pro $1.25 $0.13 $10.00 none in record
gemini-2.5-flash-lite-preview $0.10 $0.01 $0.40 none in record
gemini-robotics-er-2 $2.00 $0.20 $10.00 none in record
gemini-robotics-er-2-streaming $2.00 — $10.00 none in record
gemini-2.5-computer-use-preview-10-2025 $1.25 — $10.00 none in record

DeepSeekstandard price

ModelInputCachedOutput30dLast reprice
deepseek-v4-flash $0.44 $0.01 $1.32 2026-08-17 (371%)
deepseek-v4-pro $1.32 $0.04 $3.96 2026-08-17 (355%)
deepseek-v4-flash-vision-exp $0.44 $0.01 $1.32 — none in record

Moonshotper-model pages

ModelInputCachedOutput30dLast reprice
kimi-k3 $3.00 $0.30 $15.00 none in record
kimi-k2.7-code $0.95 $0.19 $4.00 none in record
kimi-k2.7-code-highspeed $1.90 $0.38 $8.00 none in record
moonshot-v1-128k $2.00 — $5.00 none in record
moonshot-v1-128k-vision-preview $2.00 — $5.00 none in record

xAImodel catalogue

ModelInputCachedOutput30dLast reprice
grok-4.20-0309-non-reasoning $1.25 $0.20 $2.50 none in record
grok-4.20-multi-agent-0309 $1.25 $0.20 $2.50 none in record
grok-4.20-0309-reasoning $1.25 $0.20 $2.50 none in record
grok-build-0.1 $1.00 $0.20 $2.00 none in record
grok-4.6 $2.00 $0.50 $6.00 none in record

Metastandard tier

ModelInputCachedOutput30dLast reprice
muse-spark-1.2 $1.25 $0.15 $4.25 none in record

Alibabainternational storefront

ModelInputCachedOutput30dLast reprice
qwen3.7-max $2.50 — $7.50 none in record
qwen3.7-max-preview $2.50 — $7.50 none in record
qwen3.7-plus $0.40 — $1.60 none in record
qwen-plus-latest $0.40 — $1.20 none in record
qwen-turbo $0.05 — $0.20 none in record

Z.AImodel table

ModelInputCachedOutput30dLast reprice
GLM-5.3 $1.40 $0.26 $4.40 none in record
GLM-5-Turbo $1.20 $0.24 $4.00 none in record
GLM-4.7-FlashX $0.07 $0.01 $0.40 none in record
GLM-4.5-X $2.20 $0.45 $8.90 none in record
GLM-4.5-Air $0.20 $0.03 $1.10 none in record
GLM-4.5-AirX $1.10 $0.22 $4.50 none in record
GLM-4-32B-0414-128K $0.10 — $0.10 none in record
GLM-5V-Turbo $1.20 $0.24 $4.00 none in record
GLM-4.6V $0.30 $0.05 $0.90 none in record
GLM-OCR $0.03 — $0.03 none in record
GLM-4.6V-FlashX $0.04 $0.00 $0.40 none in record

MiniMaxstandard tier

ModelInputCachedOutput30dLast reprice
MiniMax-M3 $0.30 $0.06 $1.20 none in record
MiniMax-M2.7-highspeed $0.60 $0.06 $2.40 none in record

Thinking Machinesserverless inference

ModelInputCachedOutput30dLast reprice
Inkling-Small $0.30 $0.06 $1.20 none in record
Inkling $1.00 $0.17 $4.05 none in record

Xiaomioverseas price list

ModelInputCachedOutput30dLast reprice
mimo-v2.5-pro $0.43 $0.00 $0.87 none in record
mimo-v2.5 $0.14 $0.00 $0.28 none in record

Scope: these are standard self-serve rates, the price a buyer pays without a contract. Batch, provisioned-throughput and committed-spend tiers are separate prices and are not pooled into these figures, so a large buyer's realized rate sits below them. Cached input is shown wherever the owner posts it.

gpt-5.6 output price by tier · log scale · 2026-07-27 → 2026-08-22
$1 $2 $5 $10 $30 07-2708-0908-22 sol $20 terra $12 luna $1.2

One vendor, one day, three tiers: on 2026-07-30 the posted output price fell 80% on luna and 20% on terra while sol held.

02

Open-weight models — by serving breadth

cross-provider medians · n ≥ 3 · 2026-08-22
ModelProviders Input medianOutput median Output range
MiniMax M3 6 $0.30 $1.20 $0.96 – $2.40
DeepSeek V4-Pro 5 $1.32 $3.20 $2.55 – $3.96
DeepSeek V4-Flash 5 $0.14 $0.28 $0.18 – $1.32
gemma-4-31B-it 4 $0.26 $0.65 $0.38 – $1.15
gpt-oss-120b 4 $0.10 $0.42 $0.17 – $0.60
kimi-k3 4 $3.00 $15.00 $14.25 – $15.00
GLM-5 4 $1.00 $3.20 $2.08 – $3.20
DeepSeek V3.1 4 $0.41 $1.32 $0.95 – $4.50
MiniMax-M2 4 $0.30 $1.20 $1.02 – $1.20
MiniMax M2.5 4 $0.30 $1.20 $1.15 – $1.20
MiniMax-M2.7 4 $0.30 $1.20 $1.00 – $2.40
DeepSeek V4-Flash-0731 4 $0.17 $0.47 $0.18 – $1.32
Kimi K2 Instruct 3 $0.57 $2.30 $2.00 – $2.50
Qwen3 Coder 480B A35B Instruct 3 $0.40 $1.60 $1.55 – $1.80
DeepSeek-V3.2 3 $0.27 $0.40 $0.38 – $4.50
qwen3.7-max Currently equivalent to qwen3.7-max-2026-05-20 context caching discount 3 $2.50 $7.50 $3.75 – $7.50

Each row is a median across independent providers of the same model, n disclosed. A wide range is dispersion in the serving market.

Some models sit in both sections: the owner posts a price and independent providers also serve the open weights. The two numbers are different measurements, not a restatement. deepseek-v4-flash is posted at $1.32 per 1M output tokens above, and the median of independent providers here is $0.28. What a model owner asks and what the serving market charges are separate facts, and neither is pooled into the other.

Method
  • USD per 1 million tokens, as posted. Record runs from 2026-07-03.
  • Posted rows come from the model owners' own price lists, captured daily; a reprice appears on the day the list changed.
  • Served rows are medians across independent providers of the same open-weight model, published at n ≥ 3.
  • The two are never pooled. A posted price and a cross-provider median answer different questions.
  • Token prices are not evidence about GPU rental rates. The link between them is serving efficiency, a vendor variable rather than a market price. CCIR publishes the GPU-hour record separately.