AI API pricing per million tokens, by vendor

Notes on these rows

GPT-5.6 Luna, GPT-5.4 nano, GPT-5.4 mini, GPT-5.6 Terra, GPT-5.6 Sol

Standard tier. Context and max output from the model page (second source).

GPT-6 Astra

Standard tier. Context and max output are printed on the model page (second source) as "1,050,000 context window" and "128,000 max output tokens", not on the pricing table itself. Regional (data residency) endpoints carry a 10% uplift for models released on or after 5 Mar 2026.

Notes on these rows

Claude Haiku 4.5

Alias claude-haiku-4-5. Cache writes $1.25 (5 min) / $2 (1 hour). Context 200K and max output 64K are from the models overview page, not the pricing page.

Claude Sonnet 5

Pricing page note: the $2/$10 introductory price "is now the standard price"; the increase to $3/$15 scheduled for 1 Sep 2026 "will not occur". Cache writes $2.50 (5 min) / $4 (1 hour). Max output from the models overview page.

Claude Opus 5

Cache writes: $6.25/MTok (5 min), $10/MTok (1 hour). Batch API 50% off. Fast mode (research preview) is $10 in / $50 out. Max output from the models overview page.

Claude Fable 5.1

Cached input = "Cache hits and refreshes" ($0.25/MTok, 0.025x base on this model). Cache writes: $12.50/MTok (5 min), $20/MTok (1 hour). Batch API 50% off. Pricing page states Claude 4.6 and later include the full 1M token context at standard pricing; the 128K max output is from the models overview page (platform.claude.com/docs/en/about-claude/models/overview).

Notes on these rows

Gemini 3.5 Flash-Lite

Paid tier. Token limits from the model page ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite.

Gemini 2.5 Flash

Paid tier. The pricing page itself notes a "1M token" context for this model; exact limits from the model page ai.google.dev/gemini-api/docs/models/gemini-2.5-flash.

Gemini 3.8 Flash

Paid tier. Promotional: "$0.75 through December 31, 2026. $1.50 starting January 1, 2027" (output $3.75 then $7.50; caching $0.075 then $0.15). Token limits from the model page ai.google.dev/gemini-api/docs/models/gemini-3.8-flash ("1,048,576" input, "65,536" output). No second official page confirming the price was reachable.

Gemini 2.5 Pro

Paid tier, prompts of 200k tokens or less. Prompts over 200k: input $2.50, output $15.00, caching $0.25. Token limits from the model page ai.google.dev/gemini-api/docs/models/gemini-2.5-pro.

Gemini 3.5 Flash

Paid tier. Token limits from the model page ai.google.dev/gemini-api/docs/models/gemini-3.5-flash.

Gemini 3.1 Pro Preview

Paid tier, prompts of 200k tokens or less. Prompts over 200k: input $4.00, output $18.00, caching $0.40. Token limits from the model page ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview.

Notes on these rows

Grok Build 0.1

Aliases grok-code-fast-1, grok-code-fast, grok-code-fast-1-0825. Prompts under 200k tokens; 200k or more: $2.00 / $0.40 cached / $4.00. Context printed as "256k" on the table and "256,000 tokens" on the model page. No max output published.

Grok 4.3

Alias grok-4.3-latest. Prompts under 200k tokens; 200k or more: $2.50 / $0.40 cached / $5.00. Model page mentions a 20% Batch API discount. Context printed as "1M" on the table and "1,000,000 tokens" on the model page. No max output published.

Grok 4.6

Prompts under 200k tokens. Prompts of 200k or more: input $4.00, cached $1.00, output $12.00; "Requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request." Context printed as "500k" on the table and "500,000 tokens" on the model page. No max output published.

Grok 4.5

Aliases grok-4.5-latest, grok-build-latest. Prompts under 200k tokens; 200k or more: $4.00 / $0.60 cached / $12.00. No max output published.

Notes on these rows

DeepSeek V4 Flash

Peak-hour prices; the table prints PEAK and OFF-PEAK rows for every price. Off-peak: cache hit $0.007, cache miss $0.22, output $0.66 per 1M tokens. Vendor: "Off-peak rates are half of the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)." Input is the "1M INPUT TOKENS (CACHE MISS)" row and cached input the "(CACHE HIT)" row. Context printed as "1M" and max output as "MAXIMUM: 384K". Model version DeepSeek-V4-Flash-0731. The Chinese-language page for the same platform (api-docs.deepseek.com/zh-cn/quick_start/pricing/) prints the peak prices in yuan: 0.10 / 3.0 / 9.0 yuan per 1M tokens for cache hit / cache miss / output, off-peak half, with Beijing-time peak hours.

DeepSeek V4 Flash Vision (experimental)

Peak-hour prices; the table prints PEAK and OFF-PEAK rows for every price. Off-peak: cache hit $0.007, cache miss $0.22, output $0.66 per 1M tokens. Vendor: "Off-peak rates are half of the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)." Input is the "1M INPUT TOKENS (CACHE MISS)" row and cached input the "(CACHE HIT)" row. Context printed as "1M" and max output as "MAXIMUM: 384K". Model version DeepSeek-V4-Flash-Vision-Exp. The Chinese-language page for the same platform (api-docs.deepseek.com/zh-cn/quick_start/pricing/) prints the peak prices in yuan: 0.10 / 3.0 / 9.0 yuan per 1M tokens for cache hit / cache miss / output, off-peak half, with Beijing-time peak hours. Vendor: "Images sent to deepseek-v4-flash-vision-exp are converted into tokens based on their dimensions and billed as input tokens together with your text tokens."

DeepSeek V4 Pro

Peak-hour prices; the table prints PEAK and OFF-PEAK rows for every price. Off-peak: cache hit $0.022, cache miss $0.66, output $1.98 per 1M tokens. Vendor: "Off-peak rates are half of the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)." Input is the "1M INPUT TOKENS (CACHE MISS)" row and cached input the "(CACHE HIT)" row. Context printed as "1M" and max output as "MAXIMUM: 384K". Model version DeepSeek-V4-Pro-0813. The Chinese-language page for the same platform (api-docs.deepseek.com/zh-cn/quick_start/pricing/) prints the peak prices in yuan: 0.30 / 9.0 / 27.0 yuan per 1M tokens for cache hit / cache miss / output, off-peak half, with Beijing-time peak hours.

Notes on these rows

Kimi K2.7 Code, Kimi K2.6, Kimi K2.7 Code (high speed)

Context printed as "262,144 tokens". Input is the "Input Price (Cache Miss)" column and cached input the "Input Price (Cache Hit)" column, per 1M tokens ("1M = 1,000,000"). No maximum output is published on the pricing page or on the model list (platform.kimi.ai/docs/models). "Prices exclude applicable taxes." platform.moonshot.ai redirects here.

Kimi K3

Context printed as "1,048,576 tokens". Input is the "Input Price (Cache Miss)" column and cached input the "Input Price (Cache Hit)" column, per 1M tokens ("1M = 1,000,000"). No maximum output is published on the pricing page or on the model list (platform.kimi.ai/docs/models). "Prices exclude applicable taxes." platform.moonshot.ai redirects here.

Notes on these rows

Kimi K2.7 Code, Kimi K2.6, Kimi K2.7 Code (high speed)

Context printed as "262,144 tokens". China platform (platform.kimi.com; platform.moonshot.cn redirects here), priced in yuan per 1M tokens: 缓存未命中 (cache miss) is the input price, 缓存命中 (cache hit) the cached input price, 输出 the output price. No maximum output is published on the pricing page or on the model list (platform.kimi.com/docs/models).

Kimi K3

Context printed as "1,048,576 tokens". China platform (platform.kimi.com; platform.moonshot.cn redirects here), priced in yuan per 1M tokens: 缓存未命中 (cache miss) is the input price, 缓存命中 (cache hit) the cached input price, 输出 the output price. No maximum output is published on the pricing page or on the model list (platform.kimi.com/docs/models).

Notes on these rows

GLM-5.3-Flash

Promotional price. Vendor: "GLM-5.3-Flash is available at a 50% discount (strikethrough prices are list prices). The promotion ends at 24:00 on September 9, 2026 (UTC+8, Singapore time)." List prices: input $0.15, cached input $0.03, output $0.50. Cached input storage is "Limited-time Free". Context "1M" and max output "128K" from the model page docs.z.ai/guides/llm/glm-5.3-flash.

GLM-4.7

Cached input storage is "Limited-time Free". Context "200K" and max output "128K" from the model page docs.z.ai/guides/llm/glm-4.7.

GLM-5

Cached input storage is "Limited-time Free". Context "200K" and max output "128K" from the model page docs.z.ai/guides/llm/glm-5.

GLM-5.3

Cached input storage is "Limited-time Free". Context "1M" and max output "128K" from the model page docs.z.ai/guides/llm/glm-5.3.

GLM-5.2

Cached input storage is "Limited-time Free". Context "1M" and max output "128K" from the model page docs.z.ai/guides/llm/glm-5.2.

GLM-5.1

Cached input storage is "Limited-time Free". Context "200K" and max output "128K" from the model page docs.z.ai/guides/llm/glm-5.1.

Notes on these rows

GLM-5.3-Flash

Promotional price. The row is marked 5折限时两周 (50% off, limited to two weeks; no end date printed) with list prices struck through: 0.8元 input, 2.8元 output, 0.23元 cache hit. China platform (open.bigmodel.cn), priced in yuan (元) per 1M tokens: 输入单价 is the input price, 输出单价 the output price, 缓存命中 the cached input price; 缓存存储 (cache storage) is 限时免费 (limited-time free). The page is JavaScript only and was read with a headless browser. Context "1M" printed on the pricing page; max output "128K" from docs.bigmodel.cn/cn/guide/start/model-overview.

GLM-4.7-FlashX

China platform (open.bigmodel.cn), priced in yuan (元) per 1M tokens: 输入单价 is the input price, 输出单价 the output price, 缓存命中 the cached input price; 缓存存储 (cache storage) is 限时免费 (limited-time free). The page is JavaScript only and was read with a headless browser. Context "200K" printed on the pricing page; max output "128K" from docs.bigmodel.cn/cn/guide/start/model-overview.

GLM-5

Tier is the vendor's input-length band in thousand tokens. The 输入长度 [32+) band is 6元 input, 22元 output, 1.5元 cache hit. China platform (open.bigmodel.cn), priced in yuan (元) per 1M tokens: 输入单价 is the input price, 输出单价 the output price, 缓存命中 the cached input price; 缓存存储 (cache storage) is 限时免费 (limited-time free). The page is JavaScript only and was read with a headless browser. Context "200K" from docs.bigmodel.cn/cn/guide/start/model-overview; max output "128K" from docs.bigmodel.cn/cn/guide/start/model-overview.

GLM-5.1

Tier is the vendor's input-length band in thousand tokens. The 输入长度 [32+) band is 8元 input, 28元 output, 2元 cache hit. China platform (open.bigmodel.cn), priced in yuan (元) per 1M tokens: 输入单价 is the input price, 输出单价 the output price, 缓存命中 the cached input price; 缓存存储 (cache storage) is 限时免费 (limited-time free). The page is JavaScript only and was read with a headless browser. Context "200K" from docs.bigmodel.cn/cn/guide/start/model-overview; max output "128K" from docs.bigmodel.cn/cn/guide/start/model-overview.

GLM-5.3, GLM-5.2

China platform (open.bigmodel.cn), priced in yuan (元) per 1M tokens: 输入单价 is the input price, 输出单价 the output price, 缓存命中 the cached input price; 缓存存储 (cache storage) is 限时免费 (limited-time free). The page is JavaScript only and was read with a headless browser. Context "1M" printed on the pricing page; max output "128K" from docs.bigmodel.cn/cn/guide/start/model-overview.

Notes on these rows

Qwen3.7-Flash

Singapore region, "International" deployment scope, page "Last Updated: Sep 04, 2026". Lowest band. Other bands: 32K<Token≤256K $0.100 input / $0.400 output; 256K<Token≤1M $0.200 input / $0.800 output. Currently equivalent to qwen3.7-flash-2026-07-15. Batch inference is 50% off. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.8-Flash

Singapore region, "International" deployment scope, page "Last Updated: Sep 04, 2026". Single band. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.7-Plus

Singapore region, "International" deployment scope, page "Last Updated: Sep 04, 2026". List price. Vendor prints "List price $0.4 (Limited-time 20% off)" for input and "List price $1.6 (Limited-time 20% off)" for output; the discounted figure is not printed, so it is not recorded. Second band 256K<Token≤1M: list price $1.2 input, $4.8 output, same wording. Thinking and Non-Thinking output prices are equal. Currently equivalent to qwen3.7-plus-2026-05-26. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.8-Max

Singapore region, "International" deployment scope, page "Last Updated: Sep 04, 2026". "Non-Thinking and Thinking modes"; single band. Free quota "1 million tokens" in Singapore only. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.7-Max

Singapore region, "International" deployment scope, page "Last Updated: Sep 04, 2026". Currently equivalent to qwen3.7-max-2026-05-20. "Non-Thinking and Thinking modes"; single band. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Notes on these rows

Qwen3.7-Flash

China (Beijing) region on the same international pricing page, printed in US dollars by the vendor; page "Last Updated: Sep 04, 2026". Lowest band. Other bands: 32K<Token≤256K $0.083 input / $0.330 output; 256K<Token≤1M $0.165 input / $0.660 output. Currently equivalent to qwen3.7-flash-2026-07-15. Batch inference is 50% off. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.8-Flash

China (Beijing) region on the same international pricing page, printed in US dollars by the vendor; page "Last Updated: Sep 04, 2026". Single band. Batch inference is 50% off. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.7-Plus

China (Beijing) region on the same international pricing page, printed in US dollars by the vendor; page "Last Updated: Sep 04, 2026". List price. Vendor prints "List price $0.276 (Limited-time 20% off)" for input and "List price $1.101 (Limited-time 20% off)" for output; the discounted figure is not printed, so it is not recorded. Second band 256K<Token≤1M: list price $0.826 input, $3.301 output, same wording. Thinking and Non-Thinking output prices are equal. Currently equivalent to qwen3.7-plus-2026-05-26. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.8-Max

China (Beijing) region on the same international pricing page, printed in US dollars by the vendor; page "Last Updated: Sep 04, 2026". "Non-Thinking and Thinking modes"; single band. Batch inference is 50% off. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Qwen3.7-Max

China (Beijing) region on the same international pricing page, printed in US dollars by the vendor; page "Last Updated: Sep 04, 2026". Currently equivalent to qwen3.7-max-2026-05-20. "Non-Thinking and Thinking modes"; single band. Batch inference is 50% off. "All tokens in the request are billed at the unit price of the corresponding tier." Cache-hit input tokens "are billed at unit prices different from the standard input price" under the vendor's Context Cache rules; no per-model cached price is printed on this page. Context "1M" from the text generation model page (alibabacloud.com/help/en/model-studio/text-generation-model). No max output published.

Notes on these rows

MiniMax-M3

Standard tier. The row is marked "Permanent 50% off" with list prices struck through: $0.60 input, $2.40 output, $0.12 cached; the charged prices are recorded. The "> 512k input tokens" band is $0.60 input, $2.40 output, $0.12 cached (struck-through $1.20 / $4.80 / $0.24). Priority tier "is 1.5x standard". Cached input is the "Prompt caching Read" column. Context "1,000,000" from the text generation guide (platform.minimax.io/docs/guides/text-generation.md). No max output published.

MiniMax-M2.7, MiniMax-M2.7 (high speed)

"Prompt caching Write" is $0.375 per M tokens. Cached input is the "Prompt caching Read" column. Context "204,800" from the text generation guide (platform.minimax.io/docs/guides/text-generation.md). No max output published.

MiniMax-M2.5

Listed under legacy models. "Prompt caching Write" is $0.375 per M tokens. Cached input is the "Prompt caching Read" column. Context "204,800" from the text generation guide (platform.minimax.io/docs/guides/text-generation.md). No max output published.

Notes on these rows

Ministral 3 (3B)

Cached input: the page states "Cached input tokens -90% on input token" as a rule with no per-model figure, so none is recorded. Context "256k" from the model page (second source), which prints the same prices. No max output published. Dated id ministral-3b-2512.

Mistral Small 4

Cached input: the page states "Cached input tokens -90% on input token" as a rule with no per-model figure, so none is recorded. Context "256k" from the model page (second source), which prints the same prices. No max output published. Dated id mistral-small-2603.

Ministral 3 (8B)

Cached input: the page states "Cached input tokens -90% on input token" as a rule with no per-model figure, so none is recorded. Context "256k" from the model page (second source), which prints the same prices. No max output published. Dated id ministral-8b-2512.

Codestral

Cached input: the page states "Cached input tokens -90% on input token" as a rule with no per-model figure, so none is recorded. Context "128k" from the model page (second source), which prints the same prices. No max output published. Dated id codestral-2508.

Mistral Large 3

Cached input: the page states "Cached input tokens -90% on input token" as a rule with no per-model figure, so none is recorded. Context "256k" from the model page (second source), which prints the same prices. No max output published. Dated id mistral-large-2512.

Mistral Medium 3.5

Cached input: the page states "Cached input tokens -90% on input token" as a rule with no per-model figure, so none is recorded. Context "256k" from the model page (second source), which prints the same prices. No max output published.

Notes on these rows

Command A+

The pricing card reads "Free" with "API key 0" and "Model download 0"; it is not priced per million tokens, so the row records zero. Card bullets: "Apache 2.0 license", "Runs on just two H100 GPUs". Context "128k" and max output "64k". Model id, context and max output from docs.cohere.com/docs/models. No cached input price published.

Command R7B

Card "Command R7B": "128K token context window", "4K maximum output tokens". Model id, context and max output from docs.cohere.com/docs/models. No cached input price published.

Command R (08-2024)

The card is labelled "Command R" with no dated id, "128K token context window", "4K maximum output tokens". The models page lists command-r-08-2024 as the current Command R with the same limits; the older Command R 03-2024 is priced separately in the page's legacy list at $0.50 input and $1.50 output. Model id, context and max output from docs.cohere.com/docs/models. No cached input price published.

Command R+ (08-2024)

From the page's legacy list: "Command R+ 08-2024 pricing is $2.50/1M tokens for input and $10.00/1M tokens for output". Context "128k" and max output "4k". Model id, context and max output from docs.cohere.com/docs/models. No cached input price published.