Original research · Official price pages · CC BY 4.0
State of LLM API Pricing 2026: what a support chat really costs
On this page
- 81models from 9 providers
- $2.29cheapest, per 10k support chats
- $100median, per 10k support chats
- 1,718xcheapest to priciest spread
The short answer
- A support chat costs between $0.0002 and $0.39 at list price. Same 4-turn conversation, 81 models, 9 providers: a 1,718x spread. The median model costs $100 per 10,000 conversations.
- $2 in / $10 out is the 2026 mid-tier price point. 6 models from 3 providers (Anthropic, Google, OpenAI) list exactly $2 / $10 per million tokens.
- Input is most of the bill. Because every turn resends the system prompt and history, input is 76% of the cost on a $2 / $10 model in our default conversation. Caching and shorter system prompts matter more than output length.
- Several prices change soon. Google lists Gemini 4 Carbon and Argon doubling to $4 / $20 on 1 Dec 2026, and OpenAI's gpt-5.6-sol promotional price ends on 30 Nov 2026.
Key numbers
- List prices across 81 models from 9 providers run from $0.017/$0.112 (granite-4.0-h-micro) to $30/$180 (gpt-5.5-pro) per 1M input/output tokens. (State of LLM API Pricing 2026, 12 Nov 2026)
- From 1 Dec 2026, Gemini API requests to gemini-2.5-pro route to gemini-3.8-flash and are billed at its rates. (Gemini API deprecations, 13 Nov 2026)
Price-per-million-tokens tables don't tell you what a customer conversation costs, because chat apps resend the system prompt and history every turn. We read every price from the provider's own pricing or model page on 12 Nov 2026, then priced the same support conversation on all 81 models. Change the assumptions in the calculator and the whole table updates.
Key findings
- The cheapest and priciest models are 1,718x apart. granite-4.0-h-micro (Cloudflare Workers AI) costs $2.29 per 10,000 conversations. gpt-5.5-pro (OpenAI) costs $3,930 (Cloudflare, OpenAI).
- Flagship prices fell, then started rising again. Anthropic's page lists Opus 4 / 4.1 at $15 / $75, Opus 4.5 to 5 at $5 / $25 and Opus 5.5 at $4 / $20 (Anthropic pricing). Google's page lists Gemini 4 Carbon and Argon going from $2 / $10 to $4 / $20 on 1 Dec 2026 (Gemini pricing).
- Cached-input prices are common but not universal. 9 of 81 rows list a cached-input price on the page we used. In 8 of those 9, the cached price is a tenth of the normal input price or less, so a long, stable system prompt is much cheaper to resend.
- Open-weight models are cheapest on hosted platforms. gpt-oss-120b costs $0.15 / $0.60 on Groq and Together AI and $0.35 / $0.75 on Cloudflare Workers AI (Groq, Together, Workers AI).
- Price pages don't always agree with themselves. Mistral's pricing FAQ says Mistral Large costs $0.5 / $1.5, while the Mistral Large 4 model page says $0.7 / $2.1. We use the model page and flag the gap (Mistral docs).
Cost-per-conversation calculator
Defaults: a 2,000-token system prompt, 4 turns, 60 tokens per customer message and 150 per reply. That's 9,500 input and 600 output tokens per conversation.
Price table
| Official source and notes | ||||||
|---|---|---|---|---|---|---|
gpt-5.5-proOpenAI | $30 | $180 | Not disclosed | $0.39 | $3,930 | developers.openai.comStandard tier |
gpt-6-astraOpenAI | $10 | $50 | $1 | $0.12 | $1,250 | developers.openai.comStandard tier, short context (<=272K input) |
gpt-5.5OpenAI | $5 | $30 | Not disclosed | $0.066 | $655 | developers.openai.comStandard tier |
gpt-5.6-solOpenAI | $4 | $20 | Not disclosed | $0.050 | $500 | developers.openai.comPromotional pricing available until November 30, 2026 (per page) |
gpt-5.4OpenAI | $2.50 | $15 | Not disclosed | $0.033 | $328 | developers.openai.comStandard tier |
gpt-4oOpenAI | $2.50 | $10 | Not disclosed | $0.030 | $298 | developers.openai.comStandard tier |
gpt-6.1-solOpenAI | $2 | $10 | $0.10 | $0.025 | $250 | developers.openai.comStandard tier, short context |
gpt-6-solOpenAI | $2 | $10 | $0.20 | $0.025 | $250 | developers.openai.comStandard tier, short context |
gpt-5.6-terraOpenAI | $2 | $12 | Not disclosed | $0.026 | $262 | developers.openai.comStandard tier |
gpt-4.1OpenAI | $2 | $8 | Not disclosed | $0.024 | $238 | developers.openai.comStandard tier |
o3OpenAI | $2 | $8 | Not disclosed | $0.024 | $238 | developers.openai.comStandard tier |
gpt-5.2OpenAI | $1.75 | $14 | Not disclosed | $0.025 | $250 | developers.openai.comStandard tier |
gpt-5.1OpenAI | $1.25 | $10 | Not disclosed | $0.018 | $179 | developers.openai.comStandard tier |
gpt-5OpenAI | $1.25 | $10 | Not disclosed | $0.018 | $179 | developers.openai.comStandard tier |
gpt-6.1-terraOpenAI | $1 | $5 | $0.05 | $0.013 | $125 | developers.openai.comStandard tier, short context |
gpt-5.4-miniOpenAI | $0.75 | $4.50 | Not disclosed | $0.010 | $98.25 | developers.openai.comStandard tier |
gpt-4.1-miniOpenAI | $0.40 | $1.60 | Not disclosed | $0.005 | $47.60 | developers.openai.comStandard tier |
gpt-5-miniOpenAI | $0.25 | $2 | Not disclosed | $0.004 | $35.75 | developers.openai.comStandard tier |
gpt-5.6-lunaOpenAI | $0.20 | $1.20 | Not disclosed | $0.003 | $26.20 | developers.openai.comStandard tier |
gpt-5.4-nanoOpenAI | $0.20 | $1.25 | Not disclosed | $0.003 | $26.50 | developers.openai.comStandard tier |
gpt-4o-miniOpenAI | $0.15 | $0.60 | Not disclosed | $0.002 | $17.85 | developers.openai.comStandard tier |
gpt-6-lunaOpenAI | $0.10 | $0.50 | $0.01 | $0.001 | $12.50 | developers.openai.comStandard tier, short context |
gpt-5-nanoOpenAI | $0.05 | $0.40 | Not disclosed | $0.0007 | $7.15 | developers.openai.comStandard tier |
Claude Opus 4.1Anthropic | $15 | $75 | Not disclosed | $0.19 | $1,875 | platform.claude.comListed as retired |
Claude Fable 5.1Anthropic | $10 | $50 | Not disclosed | $0.12 | $1,250 | platform.claude.comBase input / output |
Claude Fable 5.2Anthropic | $8 | $40 | Not disclosed | $0.10 | $1,000 | platform.claude.comBase input / output (same price listed for Mythos 5.2) |
Claude Opus 5Anthropic | $5 | $25 | Not disclosed | $0.062 | $625 | platform.claude.comBase input / output (same as Opus 4.5-4.8) |
Claude Opus 4.5Anthropic | $5 | $25 | Not disclosed | $0.062 | $625 | platform.claude.comBase input / output |
Claude Opus 5.5Anthropic | $4 | $20 | Not disclosed | $0.050 | $500 | platform.claude.comBase input / output |
Claude Sonnet 4.6Anthropic | $3 | $15 | Not disclosed | $0.037 | $375 | platform.claude.comBase input / output (same as Sonnet 4.5 and 4) |
Claude Sonnet 5.5Anthropic | $2 | $10 | Not disclosed | $0.025 | $250 | platform.claude.comBase input / output |
Claude Sonnet 5Anthropic | $2 | $10 | Not disclosed | $0.025 | $250 | platform.claude.comIntroductory price made standard; planned rise to $3/$15 on Sep 1, 2026 will not occur |
Claude Haiku 4.5Anthropic | $1 | $5 | Not disclosed | $0.013 | $125 | platform.claude.comBase input / output |
Claude Haiku 5.5Anthropic | $0.10 | $0.50 | Not disclosed | $0.001 | $12.50 | platform.claude.comPrompts <=100k tokens; $0.50/$2.50 above 100k |
Gemini 4 CarbonGoogle | $2 | $10 | Not disclosed | $0.025 | $250 | ai.google.devPaid tier through Nov 30, 2026; $4/$20 from Dec 1, 2026 |
Gemini 4 ArgonGoogle | $2 | $10 | Not disclosed | $0.025 | $250 | ai.google.devPaid tier through Nov 30, 2026; $4/$20 from Dec 1, 2026 |
Gemini 3.1 Pro PreviewGoogle | $2 | $12 | Not disclosed | $0.026 | $262 | ai.google.devPrompts <=200k; $4/$18 above 200k |
Gemini 4 HydrogenGoogle | $1.50 | $7.50 | Not disclosed | $0.019 | $188 | ai.google.devPaid tier |
Gemini 2.5 ProGoogle | $1.25 | $10 | Not disclosed | $0.018 | $179 | ai.google.devPrompts <=200k |
Gemini 3.8 FlashGoogle | $0.75 | $3.75 | Not disclosed | $0.009 | $93.75 | ai.google.devThrough Dec 31, 2026; $1.50/$7.50 from Jan 1, 2027 |
Gemini 3.6 FlashGoogle | $0.75 | $3.75 | Not disclosed | $0.009 | $93.75 | ai.google.devThrough Dec 31, 2026; $1.50/$7.50 from Jan 1, 2027 |
Gemini 3 Flash PreviewGoogle | $0.50 | $3 | Not disclosed | $0.007 | $65.50 | ai.google.devPaid tier |
Gemini 3.5 Flash-LiteGoogle | $0.30 | $2.50 | Not disclosed | $0.004 | $43.50 | ai.google.devPaid tier |
Gemini 2.5 FlashGoogle | $0.30 | $2.50 | Not disclosed | $0.004 | $43.50 | ai.google.devPaid tier |
Gemini 3.1 Flash-LiteGoogle | $0.25 | $1.50 | Not disclosed | $0.003 | $32.75 | ai.google.devPaid tier |
Gemini 2.5 Flash-LiteGoogle | $0.10 | $0.40 | Not disclosed | $0.001 | $11.90 | ai.google.devPaid tier |
deepseek-v4-pro (V4-Pro)DeepSeek | $1.32 | $3.96 | $0.044 | $0.015 | $149 | api-docs.deepseek.comPeak hours (cache miss); off-peak $0.66/$1.98 |
deepseek-flash (V4.1-Flash)DeepSeek | $0.30 | $1.20 | $0.006 | $0.004 | $35.70 | api-docs.deepseek.comPeak hours (cache miss); off-peak is half: $0.15/$0.60 |
grok-4.7xAI | $2 | $6 | $0.50 | $0.023 | $226 | docs.x.aiPrompts <200k tokens; prices double at 200k+ |
grok-4.6xAI | $2 | $6 | Not disclosed | $0.023 | $226 | docs.x.aiPer page |
grok-4.5xAI | $2 | $6 | Not disclosed | $0.023 | $226 | docs.x.aiPer page |
grok-4.3xAI | $1.25 | $2.50 | Not disclosed | $0.013 | $134 | docs.x.aiPer page |
grok-4.20-0309xAI | $1.25 | $2.50 | Not disclosed | $0.013 | $134 | docs.x.aiReasoning and non-reasoning |
grok-build-0.1xAI | $1 | $2 | Not disclosed | $0.011 | $107 | docs.x.aiPer page |
Mistral Large 4Mistral | $0.70 | $2.10 | $0.07 | $0.008 | $79.10 | docs.mistral.aiModel page; note mistral.ai/pricing FAQ states $0.5 in / $1.5 out for Mistral Large |
Ministral 3 14BMistral | $0.20 | $0.20 | Not disclosed | $0.002 | $20.20 | docs.mistral.aiModel page |
Mistral Small 4Mistral | $0.15 | $0.60 | Not disclosed | $0.002 | $17.85 | docs.mistral.aiModel page shows $0.15 /M and $0.6 /M; labels are tooltips, read as input then output |
Ministral 3 8BMistral | $0.15 | $0.15 | Not disclosed | $0.002 | $15.15 | docs.mistral.aiModel page |
Ministral 3 3BMistral | $0.10 | $0.10 | Not disclosed | $0.001 | $10.10 | docs.mistral.aiModel page |
kimi-k2.5Cloudflare Workers AI | $0.60 | $3 | Not disclosed | $0.007 | $75.00 | developers.cloudflare.comNeuron-based billing |
nemotron-3-120bCloudflare Workers AI | $0.50 | $1.50 | Not disclosed | $0.006 | $56.50 | developers.cloudflare.comNeuron-based billing |
mistral-small-3.1-24bCloudflare Workers AI | $0.351 | $0.555 | Not disclosed | $0.004 | $36.68 | developers.cloudflare.comNeuron-based billing |
gpt-oss-120bCloudflare Workers AI | $0.35 | $0.75 | Not disclosed | $0.004 | $37.75 | developers.cloudflare.comNeuron-based billing |
deepseek-v4.1-flash-1017Cloudflare Workers AI | $0.30 | $1.20 | Not disclosed | $0.004 | $35.70 | developers.cloudflare.comNeuron-based billing |
llama-3.3-70b-instruct-fp8-fastCloudflare Workers AI | $0.293 | $2.253 | Not disclosed | $0.004 | $41.35 | developers.cloudflare.comNeuron-based billing |
llama-4-scout-17b-16e-instructCloudflare Workers AI | $0.27 | $0.85 | Not disclosed | $0.003 | $30.75 | developers.cloudflare.comNeuron-based billing |
gpt-oss-20bCloudflare Workers AI | $0.20 | $0.30 | Not disclosed | $0.002 | $20.80 | developers.cloudflare.comNeuron-based billing |
gemma-4-26b-a4b-itCloudflare Workers AI | $0.10 | $0.30 | Not disclosed | $0.001 | $11.30 | developers.cloudflare.comNeuron-based billing |
glm-4.7-flashCloudflare Workers AI | $0.06 | $0.40 | Not disclosed | $0.0008 | $8.10 | developers.cloudflare.comNeuron-based billing |
qwen3-30b-a3b-fp8Cloudflare Workers AI | $0.051 | $0.335 | Not disclosed | $0.0007 | $6.86 | developers.cloudflare.comNeuron-based billing |
llama-3.1-8b-instruct-fp8-fastCloudflare Workers AI | $0.045 | $0.384 | Not disclosed | $0.0007 | $6.58 | developers.cloudflare.com$0.011 per 1,000 neurons; 10,000 neurons/day free |
llama-3.2-1bCloudflare Workers AI | $0.027 | $0.201 | Not disclosed | $0.0004 | $3.77 | developers.cloudflare.comNeuron-based billing |
granite-4.0-h-microCloudflare Workers AI | $0.017 | $0.112 | Not disclosed | $0.0002 | $2.29 | developers.cloudflare.comNeuron-based billing |
qwen3.8-27bGroq | $0.80 | $4 | Not disclosed | $0.010 | $100 | console.groq.comOn-demand |
gpt-oss-120bGroq | $0.15 | $0.60 | Not disclosed | $0.002 | $17.85 | console.groq.comOn-demand |
gpt-oss-20bGroq | $0.075 | $0.30 | Not disclosed | $0.0009 | $8.92 | console.groq.comOn-demand |
Kimi K3Together AI | $3 | $15 | Not disclosed | $0.037 | $375 | together.aiServerless |
Qwen3.7-MaxTogether AI | $2.50 | $7.50 | Not disclosed | $0.028 | $282 | together.aiServerless |
GLM-5.3Together AI | $1.40 | $4.40 | Not disclosed | $0.016 | $159 | together.aiServerless |
DeepSeek V4.1 FlashTogether AI | $0.30 | $1.20 | Not disclosed | $0.004 | $35.70 | together.aiServerless |
gpt-oss-120BTogether AI | $0.15 | $0.60 | Not disclosed | $0.002 | $17.85 | together.aiServerless |
Cost per 10,000 support conversations
Price changes
We only list changes the provider's own page states (scheduled increases, promotions ending, renamed tiers). We don't reconstruct history from third-party sites.
| Model | Before | After | Source | |||
|---|---|---|---|---|---|---|
| 30 Jul 2026 | OpenAI | All models (Priority processing) | Tier renamed | Priority processing | Renamed Fast mode | developers.openai.com |
| 1 Sep 2026 | Anthropic | Claude Sonnet 5 | Planned increase cancelled | $2 / $10 (introductory) | $2 / $10 (now standard; planned $3 / $15 cancelled) | platform.claude.com |
| 30 Nov 2026 | OpenAI | gpt-5.6-sol | Promotion ends | $4 / $20 (promotional) | Not disclosed on the page | developers.openai.com |
| 1 Dec 2026 | Gemini 4 Carbon / Gemini 4 Argon | Scheduled increase | $2.00 / $10.00 | $4.00 / $20.00 | ai.google.dev | |
| 1 Jan 2027 | Gemini 3.8 Flash / Gemini 3.6 Flash | Scheduled increase | $0.75 / $3.75 | $1.50 / $7.50 | ai.google.dev | |
| Ongoing | DeepSeek | deepseek-flash, deepseek-v4-pro | Time-of-day pricing | Peak price | 50% off outside peak hours (01:00-04:00 and 06:00-10:00 UTC, Mon-Fri are peak) | api-docs.deepseek.com |
Methodology
- Sources: each provider's official pricing or model page, read on 12 Nov 2026: OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral, Cloudflare Workers AI, Groq, Together AI. Mistral and some Cloudflare rows use the per-model page.
- Which price: the standard (non-batch, non-priority) list price per 1 million tokens in USD, for the short-context tier where a page lists several. Cached input is the provider's cached or cache-hit input price. Where the row has none, we show "Not disclosed".
- Selection: current text models a support team might use, as listed on each page. We left out embedding, image, audio and fine-tuning prices.
- Cost per conversation: our arithmetic (formula under the calculator). It's an estimate of list-price cost, not a bill.
Provider notes from the official pages:
- OpenAI: Batch API is 50% off. Models released on or after March 5, 2026 add a 10% uplift for regional processing.
- Anthropic: Batch is 50% off. Claude 4.7 and later use a tokenizer that produces approximately 30% more tokens for the same text.
- Google: Output prices include thinking tokens.
- Mistral: mistral.ai/pricing FAQ: 'Mistral Large costs $0.5 /M tokens in and $1.5 /M tokens out', which differs from the Mistral Large 4 model page ($0.7 / $2.1). We use the model page and flag the gap.
Limitations
- Tokens aren't equal across providers. Different tokenizers turn the same text into different token counts. Anthropic's page says Claude 4.7 and later produce about 30% more tokens for the same text. Our calculator uses one token count for every model.
- Reasoning models bill hidden tokens. Thinking tokens are billed as output (Google says so on its page), so real output can be far above our 150 tokens per reply.
- Discounts and fees vary. Batch, priority or fast tiers, regional uplifts, long-context tiers, free allowances (for example Cloudflare's 10,000 free neurons a day) and volume deals aren't modeled.
- Prices change. This is a snapshot dated 12 Nov 2026. Check the linked page before you buy.
Download the data
Licensed under CC BY 4.0. Credit "Litmiq, State of LLM API Pricing 2026" and link to this page.
- llm-api-pricing-2026.csv - one row per model: prices, cached price (empty = not disclosed), notes, source URL, cost per conversation
- llm-api-pricing-2026.json - the same plus assumptions, provider notes and price changes
- source-verification-2026-11-12.json - URL check log
Cite this
Text
Litmiq (2026). State of LLM API Pricing 2026 [Data set]. Sources read 12 Nov 2026. https://litmiq.com/llm-api-pricing. License: CC BY 4.0.BibTeX
@misc{litmiq_llm_api_pricing_2026,
title = {State of LLM API Pricing 2026},
author = {{Litmiq}},
year = {2026},
month = nov,
note = {Official list prices for 81 LLM API models from 9 providers. Sources read 2026-11-12. License: CC BY 4.0},
howpublished = {\url{https://litmiq.com/llm-api-pricing}}
}Link
<a href="https://litmiq.com/llm-api-pricing">State of LLM API Pricing 2026</a>FAQ
What is the cheapest LLM API in 2026?
Among the 81 models we read on 12 Nov 2026, the lowest list price for our 4-turn support conversation was granite-4.0-h-micro on Cloudflare Workers AI, at $2.29 per 10,000 conversations. Price isn't quality, though: run your own test cases before you switch.
How much does an AI customer support conversation cost?
With our default assumptions (2,000-token system prompt, 4 turns, 9,500 input and 600 output tokens), list-price cost ranges from $0.0002 to $0.39 per conversation across 81 models. The median is $0.010. Use the calculator with your own token counts.
Why does input cost more than output in a chatbot?
Every turn resends the system prompt and the conversation so far, so input tokens add up fast. On a $2 / $10 model, input is 76% of the bill in our default conversation, even though output is 5 times the price per token. Shorter system prompts and prompt caching cut cost the most.
Are these prices official?
Yes. Each row links to the provider's own pricing or model page, read on 12 Nov 2026. Where a page does not list a cached-input price, we show "Not disclosed", not zero. Prices change; check the source before you buy.
Which LLM prices are changing soon?
Google's pricing page lists increases for Gemini 4 Carbon and Argon (to $4 / $20 from 1 Dec 2026) and Gemini 3.8 / 3.6 Flash (to $1.50 / $7.50 from 1 Jan 2027). OpenAI's page says gpt-5.6-sol promotional pricing ends 30 Nov 2026.
Can I reuse this data?
Yes, under CC BY 4.0. Credit "Litmiq, State of LLM API Pricing 2026" and link to this page.
Get model deprecation alerts
We email you when OpenAI, Anthropic or Google publish a new shutdown date for a model on our tracker. Sent by hand from hello@litmiq.com, a few times a month at most. Unsubscribe by replying.
We store only your email, through our form provider. See the privacy policy.
Test the cheaper model before you switch
Run your own support prompt on real test cases and compare replies side by side. Free, in your browser.
- Free, no signup
- Runs in your browser
- Your OpenAI key, sent only to OpenAI