AI API Pricing per 1 Million Tokens
Current API rates normalized to USD per 1 million tokens. Input, cached input and output are shown separately when providers price them differently.
How API token cost is calculated
Multiply the number of tokens by the provider rate, then divide by 1,000,000.
cost = tokens × price per 1M ÷ 1,000,000
Example: 250,000 input tokens at $2 / 1M = $0.50.
| Model | Rate | Price / 1M tokens |
|---|---|---|
| GPT-6 Astra OpenAI's most capable model for complex reasoning, coding, computer use, research and end-to-end professional work. 1,050,000 context · 128,000 max output · Apr 30, 2026 knowledge cutoff · text/image input · reasoning effort: low, medium, high, xhigh, max. Standard pricing is $10 input, $1 cached input, $12.50 cache write and $50 output per 1M tokens. Requests with more than 272K input tokens are billed at 2x input/cache rates and 1.5x output for the full request. Batch and Flex are 50% of Standard; Fast mode is 2x. Astra is rolling out first to limited organizations, with broader API and paid ChatGPT access over the coming days. | Input — Standard | $10 |
| GPT-6 Astra | Cached input — Standard | $1 |
| GPT-6 Astra | Cache write — Standard | $12.5 |
| GPT-6 Astra | Output — Standard | $50 |
| GPT-6.1 Sol OpenAI reasoning model with near-Astra performance for complex coding, computer use and professional work at a lower cost. 1.05M context · 128K max output · text/image input · reasoning low/medium/high/xhigh/max. Standard pricing. Requests with more than 272K input tokens are billed at 2x input/cache rates and 1.5x output for the full request. Batch and Flex are 50% lower than Standard; Fast mode is 2x. | Input — Standard, ≤272K input | $2 |
| GPT-6.1 Sol | Cached input — Standard, ≤272K | $0.1 |
| GPT-6.1 Sol | Cache write — Standard, ≤272K | $2.5 |
| GPT-6.1 Sol | Output — Standard, ≤272K input | $10 |
| GPT-6.1 Sol | Input — Standard, >272K input | $4 |
| GPT-6.1 Sol | Cached input — Standard, >272K | $0.2 |
| GPT-6.1 Sol | Cache write — Standard, >272K | $5 |
| GPT-6.1 Sol | Output — Standard, >272K input | $15 |
| GPT-6 Sol OpenAI reasoning model for complex coding and agentic workflows. 1.05M context · 128K max output · text/image input · reasoning none/low/medium/high/xhigh/max. Standard pricing. OpenAI applies the long-context tier above 272K input tokens. | Input — Standard, ≤272K input | $2 |
| GPT-6 Sol | Cached input — Standard, ≤272K | $0.2 |
| GPT-6 Sol | Cache write — Standard, ≤272K | $2.5 |
| GPT-6 Sol | Output — Standard, ≤272K input | $10 |
| GPT-6 Sol | Input — Standard, >272K input | $4 |
| GPT-6 Sol | Cached input — Standard, >272K | $0.4 |
| GPT-6 Sol | Cache write — Standard, >272K | $5 |
| GPT-6 Sol | Output — Standard, >272K input | $15 |
| GPT-6 Luna OpenAI's efficient GPT-6 model for focused, high-volume tasks. 1.05M context · 128K max output · text/image input · reasoning none/low/medium/high/xhigh/max. Standard pricing. OpenAI applies the long-context tier above 272K input tokens. | Input — Standard, ≤272K input | $0.1 |
| GPT-6 Luna | Cached input — Standard, ≤272K | $0.01 |
| GPT-6 Luna | Cache write — Standard, ≤272K | $0.125 |
| GPT-6 Luna | Output — Standard, ≤272K input | $0.5 |
| GPT-6 Luna | Input — Standard, >272K input | $0.2 |
| GPT-6 Luna | Cached input — Standard, >272K | $0.02 |
| GPT-6 Luna | Cache write — Standard, >272K | $0.25 |
| GPT-6 Luna | Output — Standard, >272K input | $0.75 |
| GPT-5.6 Sol OpenAI's flagship GPT-5.6 model for complex professional work, reasoning and coding. 1.05M context · 128K max output · text/image input · reasoning and tools. Promotional API pricing is $4 input, $0.40 cached input and $20 output per 1M tokens, available at least through November 21, 2026. Requests above 272K input tokens cost 2x input and 1.5x output for the full request; cache writes cost 1.25x uncached input. | Input — promotional pricing | $4 |
| GPT-5.6 Sol | Cached input — promotional pricing | $0.4 |
| GPT-5.6 Sol | Output — promotional pricing | $20 |
| GPT-5.6 Terra Balanced GPT-5.6 tier for strong intelligence at lower cost than Sol. 1.05M context · 128K max output · text/image input · reasoning and tools. Requests above 272K input tokens cost 2x input and 1.5x output for the full request; cache writes cost 1.25x uncached input. | Input | $2 |
| GPT-5.6 Terra | Cached input | $0.2 |
| GPT-5.6 Terra | Output | $12 |
| GPT-5.6 Luna Cost-optimized GPT-5.6 tier for high-volume workloads. 1.05M context · 128K max output · text/image input · reasoning and tools. Requests above 272K input tokens cost 2x input and 1.5x output for the full request; cache writes cost 1.25x uncached input. | Input | $0.2 |
| GPT-5.6 Luna | Cached input | $0.02 |
| GPT-5.6 Luna | Output | $1.2 |
| GPT-5.5 Previous flagship generation retained for comparison and existing integrations. Previous-generation GPT model; prefer GPT-5.6 for new integrations. | Input | $5 |
| GPT-5.5 | Cached input | $0.5 |
| GPT-5.5 | Output | $30 |
| GPT-5.4 Earlier lower-cost GPT tier retained for price history and compatibility. Previous-generation model; compare with GPT-5.6 Terra for new workloads. | Input | $2.5 |
| GPT-5.4 | Cached input | $0.25 |
| GPT-5.4 | Output | $15 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| Claude Fable 5.1 Anthropic's latest Fable model for demanding reasoning, long-horizon agentic work and complex coding. 1M context · 128K max output · adaptive thinking always on · text/image input. Input, output and cache-write rates are unchanged from Fable 5. Cache reads are 75% cheaper at $0.25 per 1M tokens, which Anthropic estimates reduces typical token-billed workloads by about 25%. | Input | $10 |
| Claude Fable 5.1 | 5m cache write | $12.5 |
| Claude Fable 5.1 | 1h cache write | $20 |
| Claude Fable 5.1 | Cache read / hit & refresh | $0.25 |
| Claude Fable 5.1 | Output | $50 |
| Claude Opus 5.5 Anthropic's latest Opus model for long-running agents, coding and professional work. Available on the Claude Platform as claude-opus-5-5; Fast mode is priced separately. Standard pricing per million tokens is $4 input, $5 for a 5-minute cache write, $8 for a 1-hour cache write, $0.20 cache reads and $20 output. Fast mode costs $8 input and $40 output. | Input | $4 |
| Claude Opus 5.5 | Cache write — 5 minute TTL | $5 |
| Claude Opus 5.5 | Cache write — 1 hour TTL | $8 |
| Claude Opus 5.5 | Cache read | $0.2 |
| Claude Opus 5.5 | Output | $20 |
| Claude Opus 5.5 | Input — Fast mode | $8 |
| Claude Opus 5.5 | Output — Fast mode | $40 |
| Claude Fable 5 Previous Fable generation retained for comparison and existing integrations. 1M context · 128K max output · previous Fable generation; Fable 5.1 is the current upgrade. | Input | $10 |
| Claude Fable 5 | Output | $50 |
| Claude Opus 5 Top Opus tier for complex agentic coding and enterprise work at half Fable's token price. 1M context · 128K max output · thinking on by default. | Input | $5 |
| Claude Opus 5 | Output | $25 |
| Claude Sonnet 5 High-performance Claude tier for coding and agents at permanent $2/$10 API pricing. 1M context · 128K max output · $2 input / $10 output per 1M tokens. Anthropic made the introductory $2 input / $10 output per 1M tokens pricing permanent on August 10, 2026; the previously planned $3/$15 pricing no longer applies. | Input | $2 |
| Claude Sonnet 5 | Output | $10 |
| Claude Haiku 4.5 Anthropic's fastest, most cost-efficient Claude tier for scaled and latency-sensitive workloads. Fast small-model tier · text/image understanding · prompt caching and batch discounts. | Input | $1 |
| Claude Haiku 4.5 | Output | $5 |
| Claude Opus 4.8 Previous Opus generation retained for comparison and existing integrations. Previous-generation Opus; Opus 5 is the current upgrade at the same token price. | Input | $5 |
| Claude Opus 4.8 | Output | $25 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| Grok 4.5 xAI's current flagship Grok model with strong reasoning and multimodal input. 500K context · text/image input · reasoning, tools and structured output. For requests at or above 200K context, xAI lists higher long-context rates; rows marked long-context show those prices. | Input — <200K context | $2 |
| Grok 4.5 | Cached input — <200K | $0.3 |
| Grok 4.5 | Output — <200K | $6 |
| Grok 4.5 | Input — ≥200K context | $4 |
| Grok 4.5 | Cached input — ≥200K | $0.6 |
| Grok 4.5 | Output — ≥200K | $12 |
| Grok 4.20 High-performance Grok model with 1M context and agentic tool calling. 1M context · text/image input · reasoning/non-reasoning variants · function calling. For requests at or above 200K context, xAI lists higher long-context rates; rows marked long-context show those prices. | Input — <200K context | $1.25 |
| Grok 4.20 | Cached input — <200K | $0.2 |
| Grok 4.20 | Output — <200K | $2.5 |
| Grok 4.20 | Input — ≥200K context | $2.5 |
| Grok 4.20 | Cached input — ≥200K | $0.4 |
| Grok 4.20 | Output — ≥200K | $5 |
| Grok Build 0.1 xAI coding model for agentic software engineering and workflow tasks. 256K context · text/image input · reasoning, tools and structured output. For requests at or above 200K context, xAI lists higher long-context rates; rows marked long-context show those prices. | Input — <200K context | $1 |
| Grok Build 0.1 | Cached input — <200K | $0.2 |
| Grok Build 0.1 | Output — <200K | $2 |
| Grok Build 0.1 | Input — ≥200K context | $2 |
| Grok Build 0.1 | Cached input — ≥200K | $0.4 |
| Grok Build 0.1 | Output — ≥200K | $4 |
| Grok 4.6 xAI's latest recommended Grok model for coding, chat and agentic workloads. 500K context · short-context and ≥200K long-context pricing tiers. xAI charges the higher long-context tier when context is at or above 200K tokens. | Input — <200K context | $2 |
| Grok 4.6 | Cached input — <200K | $0.5 |
| Grok 4.6 | Output — <200K | $6 |
| Grok 4.6 | Input — ≥200K context | $4 |
| Grok 4.6 | Cached input — ≥200K | $1 |
| Grok 4.6 | Output — ≥200K | $12 |
| Grok 4.7 xAI's latest frontier Grok model for coding, agentic tasks and knowledge work. 500K context · text/image input · no text output limit · reasoning low/medium/high/xhigh. xAI charges the higher long-context tier when prompt tokens reach 200K or more. Grok 4.7 Fast is a separate Cursor/Grok Build-only variant billed at twice these token rates. | Input — <200K context | $2 |
| Grok 4.7 | Cached input — <200K | $0.5 |
| Grok 4.7 | Output — <200K | $6 |
| Grok 4.7 | Input — ≥200K context | $4 |
| Grok 4.7 | Cached input — ≥200K | $1 |
| Grok 4.7 | Output — ≥200K | $12 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| Gemini 3.6 Flash Google's most intelligent Flash model for fast agentic and multimodal workloads. 1M context · 64K max output · multimodal input · thinking and built-in tools. Introductory pricing is $0.75 input, $0.075 cached input and $3.75 output per 1M tokens through December 31, 2026; standard $1.50/$0.15/$7.50 pricing starts January 1, 2027. | Input — promo through Dec 31, 2026 | $0.75 |
| Gemini 3.6 Flash | Context cache input — promo through Dec 31, 2026 | $0.075 |
| Gemini 3.6 Flash | Output incl. thinking — promo through Dec 31, 2026 | $3.75 |
| Gemini 3.5 Flash-Lite Google's lowest-cost GA Gemini for high-volume agentic tasks, translation and simple processing. 1M context · 64K max output · text/image/video/audio input · thinking. | Input | $0.3 |
| Gemini 3.5 Flash-Lite | Context cache input | $0.03 |
| Gemini 3.5 Flash-Lite | Output incl. thinking | $2.5 |
| Gemini 3.7 Flash Google's newest GA Flash model for coding, agents and multimodal reasoning. 1,048,576 context · 65,536 max output · text/image/video/audio/PDF input · thinking and tools. Introductory pricing runs through December 31, 2026; standard $1.50 input, $0.15 cached input and $7.50 output pricing starts January 1, 2027. | Input — promo through Dec 31, 2026 | $0.75 |
| Gemini 3.7 Flash | Context cache input — promo through Dec 31, 2026 | $0.075 |
| Gemini 3.7 Flash | Output incl. thinking — promo through Dec 31, 2026 | $3.75 |
| Gemini 3.8 Flash Google's most intelligent GA Flash model for long-horizon software engineering, autonomous agents and complex enterprise workflows. 1,048,576 context · 65,536 max output · multimodal input · tunable thinking and built-in tools. Introductory pricing runs through December 31, 2026; standard $1.50 input, $0.15 cached input and $7.50 output pricing starts January 1, 2027. | Input — promo through Dec 31, 2026 | $0.75 |
| Gemini 3.8 Flash | Context cache input — promo through Dec 31, 2026 | $0.075 |
| Gemini 3.8 Flash | Output incl. thinking — promo through Dec 31, 2026 | $3.75 |
| Gemini 3.5 Transcribe Google GA speech-to-text model for audio files with 85+ languages, speaker diarization, word-level timestamps, smart transcription and custom vocabulary. Audio input up to 1 hour per request; up to 30 minutes with diarization or word timestamps. Free tier available. The Gemini Developer API Free Tier is free of charge. Standard paid pricing is $2/M audio input tokens and $12/M text output tokens, approximately $0.005 per minute blended. | Audio input — Standard paid tier | $2 |
| Gemini 3.5 Transcribe | Text output — Standard paid tier | $12 |
| Gemini 3.5 Transcribe Live Google GA low-latency streaming speech-to-text model over WebSockets with automatic language detection and custom vocabulary. Live streaming sessions up to 10 minutes; 85+ languages. Free tier available. The Gemini Developer API Free Tier is free of charge. Standard paid pricing is $3.50/M audio input tokens and $21/M text output tokens, approximately $0.009 per minute blended. | Audio input — Standard paid tier | $3.5 |
| Gemini 3.5 Transcribe Live | Text output — Standard paid tier | $21 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| DeepSeek V4.1 Flash Official DeepSeek V4.1 Flash release, a native multimodal model served by the DeepSeek API under deepseek-flash. 1M context · max output 384K · thinking/non-thinking · vision · JSON, tool calls, Responses API and Anthropic API · 552B MoE, 8B input-active / 16B output-active. Official V4.1 Flash pricing. Peak hours are 01:00–04:00 and 06:00–10:00 UTC Monday through Friday; all other hours are off-peak at half the peak rates. The prior temporary beta model ID has been superseded by deepseek-flash. | Input — cache hit, peak | $0.006 |
| DeepSeek V4.1 Flash | Input — cache miss, peak | $0.3 |
| DeepSeek V4.1 Flash | Output — peak | $1.2 |
| DeepSeek V4.1 Flash | Input — cache hit, off-peak | $0.003 |
| DeepSeek V4.1 Flash | Input — cache miss, off-peak | $0.15 |
| DeepSeek V4.1 Flash | Output — off-peak | $0.6 |
| DeepSeek V4 Flash (legacy alias) Legacy API alias. The original V4 Flash model is retired; requests to deepseek-v4-flash are served by DeepSeek V4.1 Flash and billed at the Flash price. Legacy alias → DeepSeek V4.1 Flash · 1M context · max output 384K · vision · thinking/non-thinking. Legacy alias retained for compatibility. DeepSeek states the original V4 Flash is retired; this alias now routes to V4.1 Flash and uses the V4.1 Flash peak/off-peak prices. | Input — cache hit, peak | $0.006 |
| DeepSeek V4 Flash (legacy alias) | Input — cache miss, peak | $0.3 |
| DeepSeek V4 Flash (legacy alias) | Output — peak | $1.2 |
| DeepSeek V4 Flash (legacy alias) | Input — cache hit, off-peak | $0.003 |
| DeepSeek V4 Flash (legacy alias) | Input — cache miss, off-peak | $0.15 |
| DeepSeek V4 Flash (legacy alias) | Output — off-peak | $0.6 |
| DeepSeek V4 Pro Higher-capability DeepSeek V4 model for harder reasoning, agentic coding and knowledge work. 1M context · max output 384K · thinking/non-thinking · 1.6T total / 49B active parameters. DeepSeek now applies peak/off-peak pricing. Peak hours are 01:00–04:00 and 06:00–10:00 UTC Monday through Friday; all other hours are off-peak at half the peak rates. | Input — cache hit, peak | $0.044 |
| DeepSeek V4 Pro | Input — cache miss, peak | $1.32 |
| DeepSeek V4 Pro | Output — peak | $3.96 |
| DeepSeek V4 Pro | Input — cache hit, off-peak | $0.022 |
| DeepSeek V4 Pro | Input — cache miss, off-peak | $0.66 |
| DeepSeek V4 Pro | Output — off-peak | $1.98 |
| DeepSeek V4 Flash Vision Exp (legacy alias) Legacy API alias. The experimental V4 Flash Vision model is retired; requests to deepseek-v4-flash-vision-exp are served by DeepSeek V4.1 Flash and billed at the Flash price. Legacy alias → DeepSeek V4.1 Flash · 1M context · max output 384K · native vision · thinking/non-thinking. Legacy alias retained for compatibility. DeepSeek states this model is retired; the alias now routes to V4.1 Flash and uses the V4.1 Flash peak/off-peak prices. | Input — cache hit, peak | $0.006 |
| DeepSeek V4 Flash Vision Exp (legacy alias) | Input — cache miss, peak | $0.3 |
| DeepSeek V4 Flash Vision Exp (legacy alias) | Output — peak | $1.2 |
| DeepSeek V4 Flash Vision Exp (legacy alias) | Input — cache hit, off-peak | $0.003 |
| DeepSeek V4 Flash Vision Exp (legacy alias) | Input — cache miss, off-peak | $0.15 |
| DeepSeek V4 Flash Vision Exp (legacy alias) | Output — off-peak | $0.6 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| Kimi K3 Kimi's 2.8T flagship for long-horizon coding, knowledge work and reasoning with native vision. 1,048,576-token context · native vision · always reasons · tools, JSON and structured output. | Input — cache hit | $0.3 |
| Kimi K3 | Input — cache miss | $3 |
| Kimi K3 | Output | $15 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| MiniMax M3 Open-weight frontier coding and agentic model with native image/video understanding. 1M context · native image/video input · computer use · MSA sparse-attention architecture. Listed API rates are MiniMax's direct API prices with its permanent 50% discount. Platform-specific free promotions are represented separately as offers and do not replace these direct rates. | Input — ≤512K context | $0.3 |
| MiniMax M3 | Output — ≤512K | $1.2 |
| MiniMax M3 | Cache read — ≤512K | $0.06 |
| MiniMax M3 | Input — 512K–1M context | $0.6 |
| MiniMax M3 | Output — 512K–1M | $2.4 |
| MiniMax M3 | Cache read — 512K–1M | $0.12 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| Mistral Medium 3.5 Frontier-class multimodal Mistral model optimized for agents and coding. 256K context · 128B parameters · Modified MIT open weights · tools and structured output. | Input | $1.5 |
| Mistral Medium 3.5 | Output | $7.5 |
| Mistral Small 4 Efficient hybrid model that unifies instruct, reasoning and coding. 256K context · 119B total / 6.5B active · Apache 2.0 open weights. | Input | $0.15 |
| Mistral Small 4 | Output | $0.6 |
| Mistral Large 3 General-purpose open-weight multimodal MoE model for broad production workloads. 256K context · 675B total / 41B active · Apache 2.0 open weights. | Input | $0.5 |
| Mistral Large 3 | Output | $1.5 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| Qwen3.7 Max Largest Qwen3.7 model, designed for agentic coding, office productivity and long autonomous execution. 1M context · 65,536 max output · thinking/non-thinking · text input/output · function calling. Alibaba lists these original API prices across its global regions; limited-time console promotions are excluded. | Input — original global price | $1.65 |
| Qwen3.7 Max | Input — implicit cache | $0.33 |
| Qwen3.7 Max | Output — original global price | $4.951 |
| Qwen3.8 Max Alibaba's 2.4T-parameter flagship Qwen model for coding, office productivity and long-horizon autonomous work. 1M context · 131,072 max output · text/image/video input · multimodal reasoning and tools. Rates shown are Alibaba Cloud's original global-region prices (including US Virginia, Frankfurt, Tokyo and Hong Kong). Singapore international pricing is $2 input / $6 output per 1M tokens; limited-time console promotions are excluded. | Input — global regions | $1.65 |
| Qwen3.8 Max | Input — implicit cache, global regions | $0.206 |
| Qwen3.8 Max | Output — global regions | $4.951 |
| Qwen3.8 Max | Explicit cache creation — global regions | $2.063 |
| Qwen3.8 Max | Explicit cache read — global regions | $0.137 |
| Model | Rate | Price / 1M tokens |
|---|---|---|
| MiniMax M3 (free via OpenRouter) OpenRouter free endpoint for MiniMax M3, currently priced at $0 for prompt and completion tokens. OpenRouter currently lists this endpoint as free. Free endpoints are rate limited and availability may change independently of MiniMax direct API pricing. | Input | $0 |
| MiniMax M3 (free via OpenRouter) | Output | $0 |