LLM API pricing comparison: what typical tasks cost on 15 models
At the model providers' own API prices (prices checked ), 1,000 chat replies — from $0.25 (GPT-6 Luna) to $25.00 (Claude Fable 5.1 and GPT-6 Astra); summarizing 100 documents — from $0.06 (GPT-6 Luna) to $6.10 (Claude Fable 5.1 and GPT-6 Astra); a day of coding assistant use — from $0.20 (GPT-6 Luna) to $20.00 (Claude Fable 5.1 and GPT-6 Astra). Calculated from token counts we state below; model quality is not compared. Methodology · Source: our AI model API prices
We calculate each amount from the task's token count and the price per million tokens recorded on the date shown. It is an estimate for planning, not a bill: your own prompts, answers and discounts change it.
The three tasks and our assumptions
| Task | Requests | Input tokens per request | Output tokens per request | Input tokens in total | Output tokens in total |
|---|---|---|---|---|---|
| 1,000 chat replies | 1,000 | 1,000 | 300 | 1,000,000 (≈ 750,000 words) | 300,000 (≈ 225,000 words) |
| Summarizing 100 documents | 100 | 4,100 | 400 | 410,000 (≈ 307,500 words) | 40,000 (≈ 30,000 words) |
| A day of coding assistant use | 200 | 6,000 | 800 | 1,200,000 (≈ 900,000 words) | 160,000 (≈ 120,000 words) |
- 1,000 chat replies. A chat assistant answers 1,000 messages. Each request sends about 1,000 tokens (instructions, the conversation so far and the new message) and gets a reply of about 300 tokens.
- Summarizing 100 documents. 100 documents of about 4,000 tokens each (roughly 3,000 words, six pages) are summarized one by one: each request adds about 100 tokens of instructions and gets a summary of about 400 tokens.
- A day of coding assistant use. One developer works with a coding assistant for a day: 200 requests, each sending about 6,000 tokens (the task, the relevant files and the previous steps) and getting about 800 tokens of code and explanation.
- Words: we assume 0.75 English words per token (1,000 words ≈ 1,333 tokens). Each model counts tokens its own way, and other languages usually need more tokens per word.
- Reasoning: some models bill their internal reasoning as output tokens; for example, Google's Gemini API price page, checked , says its output price includes thinking tokens. We count only the visible answer, so a model that reasons at length can cost more than shown.
- No discounts: prompt caching, volume deals and promotional credits are not applied. Batch prices are in a separate table below. Prices are in USD as OpenRouter lists them; we do not add taxes.
- Prices: the latest check of each price on its model page. A price not recorded at the latest check is “Unknown”; an older price is not reused.
1,000 chat replies
A chat assistant answers 1,000 messages. Each request sends about 1,000 tokens (instructions, the conversation so far and the new message) and gets a reply of about 300 tokens. 1,000,000 input and 300,000 output tokens in total. Calculated — sorted by the cost at the model provider's own price, lowest first; models without that price come last.
| Model | At the model provider's own price | Through OpenRouter | Prices used, per 1M tokens (input / output) |
|---|---|---|---|
| GPT-6 Luna OpenAI | $0.25 OpenAI | $0.25 | OpenAI: $0.10 / $0.50 OpenRouter: $0.10 / $0.50 |
| Qwen3.8 Flash Qwen | $0.29 Alibaba Cloud Model Studio (Qwen) (International) | $0.29 | Alibaba Cloud Model Studio (Qwen): $0.15 / $0.47 OpenRouter: $0.15 / $0.47 |
| gpt-oss-120b OpenAI | $0.33 Groq | $0.33 | Groq: $0.15 / $0.60 OpenRouter: $0.15 / $0.60 |
| DeepSeek V4.1 Flash DeepSeek | $0.66 DeepSeek (peak rate) | $0.08 $0.08–$0.23 over 7 days | DeepSeek: $0.30 / $1.20 OpenRouter: $0.032 / $0.144 |
| Mistral Large 3 2512 Mistral | $0.95 Mistral AI | $0.95 | Mistral AI: $0.50 / $1.50 OpenRouter: $0.50 / $1.50 |
| Gemini 3.8 Flash | $1.88 Google Gemini API | $2.06 | Google Gemini API: $0.75 / $3.75 OpenRouter: $0.825 / $4.125 |
| DeepSeek V4 Pro 0813 DeepSeek | $2.51 DeepSeek (peak rate) | $1.25 | DeepSeek: $1.32 / $3.96 OpenRouter: $0.66 / $1.98 |
| Qwen3.8 Max (0902) Qwen | $3.80 Alibaba Cloud Model Studio (Qwen) (International) | $3.80 | Alibaba Cloud Model Studio (Qwen): $2.00 / $6.00 OpenRouter: $2.00 / $6.00 |
| Grok 4.7 SpaceXAI | $3.80 xAI | $3.80 | xAI: $2.00 / $6.00 OpenRouter: $2.00 / $6.00 |
| Claude Sonnet 5.5 Anthropic | $5.00 Anthropic | $5.00 | Anthropic: $2.00 / $10.00 OpenRouter: $2.00 / $10.00 |
| GPT-6 Sol OpenAI | $5.00 OpenAI | $5.00 | OpenAI: $2.00 / $10.00 OpenRouter: $2.00 / $10.00 |
| Gemini 3.1 Pro Preview | $5.60 Google Gemini API | $5.60 | Google Gemini API: $2.00 / $12.00 OpenRouter: $2.00 / $12.00 |
| Claude Opus 5.5 Anthropic | $10.00 Anthropic | $10.00 | Anthropic: $4.00 / $20.00 OpenRouter: $4.00 / $20.00 |
| Claude Fable 5.1 Anthropic | $25.00 Anthropic | $25.00 | Anthropic: $10.00 / $50.00 OpenRouter: $10.00 / $50.00 |
| GPT-6 Astra OpenAI | $25.00 OpenAI | $25.00 | OpenAI: $10.00 / $50.00 OpenRouter: $10.00 / $50.00 |
Summarizing 100 documents
100 documents of about 4,000 tokens each (roughly 3,000 words, six pages) are summarized one by one: each request adds about 100 tokens of instructions and gets a summary of about 400 tokens. 410,000 input and 40,000 output tokens in total. Calculated — sorted by the cost at the model provider's own price, lowest first; models without that price come last.
| Model | At the model provider's own price | Through OpenRouter | Prices used, per 1M tokens (input / output) |
|---|---|---|---|
| GPT-6 Luna OpenAI | $0.06 OpenAI | $0.06 | OpenAI: $0.10 / $0.50 OpenRouter: $0.10 / $0.50 |
| Qwen3.8 Flash Qwen | $0.08 Alibaba Cloud Model Studio (Qwen) (International) | $0.08 | Alibaba Cloud Model Studio (Qwen): $0.15 / $0.47 OpenRouter: $0.15 / $0.47 |
| gpt-oss-120b OpenAI | $0.09 Groq | $0.09 | Groq: $0.15 / $0.60 OpenRouter: $0.15 / $0.60 |
| DeepSeek V4.1 Flash DeepSeek | $0.17 DeepSeek (peak rate) | $0.02 $0.02–$0.04 over 7 days | DeepSeek: $0.30 / $1.20 OpenRouter: $0.032 / $0.144 |
| Mistral Large 3 2512 Mistral | $0.27 Mistral AI | $0.27 | Mistral AI: $0.50 / $1.50 OpenRouter: $0.50 / $1.50 |
| Gemini 3.8 Flash | $0.46 Google Gemini API | $0.50 | Google Gemini API: $0.75 / $3.75 OpenRouter: $0.825 / $4.125 |
| DeepSeek V4 Pro 0813 DeepSeek | $0.70 DeepSeek (peak rate) | $0.35 | DeepSeek: $1.32 / $3.96 OpenRouter: $0.66 / $1.98 |
| Qwen3.8 Max (0902) Qwen | $1.06 Alibaba Cloud Model Studio (Qwen) (International) | $1.06 | Alibaba Cloud Model Studio (Qwen): $2.00 / $6.00 OpenRouter: $2.00 / $6.00 |
| Grok 4.7 SpaceXAI | $1.06 xAI | $1.06 | xAI: $2.00 / $6.00 OpenRouter: $2.00 / $6.00 |
| Claude Sonnet 5.5 Anthropic | $1.22 Anthropic | $1.22 | Anthropic: $2.00 / $10.00 OpenRouter: $2.00 / $10.00 |
| GPT-6 Sol OpenAI | $1.22 OpenAI | $1.22 | OpenAI: $2.00 / $10.00 OpenRouter: $2.00 / $10.00 |
| Gemini 3.1 Pro Preview | $1.30 Google Gemini API | $1.30 | Google Gemini API: $2.00 / $12.00 OpenRouter: $2.00 / $12.00 |
| Claude Opus 5.5 Anthropic | $2.44 Anthropic | $2.44 | Anthropic: $4.00 / $20.00 OpenRouter: $4.00 / $20.00 |
| Claude Fable 5.1 Anthropic | $6.10 Anthropic | $6.10 | Anthropic: $10.00 / $50.00 OpenRouter: $10.00 / $50.00 |
| GPT-6 Astra OpenAI | $6.10 OpenAI | $6.10 | OpenAI: $10.00 / $50.00 OpenRouter: $10.00 / $50.00 |
A day of coding assistant use
One developer works with a coding assistant for a day: 200 requests, each sending about 6,000 tokens (the task, the relevant files and the previous steps) and getting about 800 tokens of code and explanation. 1,200,000 input and 160,000 output tokens in total. Calculated — sorted by the cost at the model provider's own price, lowest first; models without that price come last.
| Model | At the model provider's own price | Through OpenRouter | Prices used, per 1M tokens (input / output) |
|---|---|---|---|
| GPT-6 Luna OpenAI | $0.20 OpenAI | $0.20 | OpenAI: $0.10 / $0.50 OpenRouter: $0.10 / $0.50 |
| Qwen3.8 Flash Qwen | $0.26 Alibaba Cloud Model Studio (Qwen) (International) | $0.26 | Alibaba Cloud Model Studio (Qwen): $0.15 / $0.47 OpenRouter: $0.15 / $0.47 |
| gpt-oss-120b OpenAI | $0.28 Groq | $0.28 | Groq: $0.15 / $0.60 OpenRouter: $0.15 / $0.60 |
| DeepSeek V4.1 Flash DeepSeek | $0.55 DeepSeek (peak rate) | $0.06 $0.06–$0.15 over 7 days | DeepSeek: $0.30 / $1.20 OpenRouter: $0.032 / $0.144 |
| Mistral Large 3 2512 Mistral | $0.84 Mistral AI | $0.84 | Mistral AI: $0.50 / $1.50 OpenRouter: $0.50 / $1.50 |
| Gemini 3.8 Flash | $1.50 Google Gemini API | $1.65 | Google Gemini API: $0.75 / $3.75 OpenRouter: $0.825 / $4.125 |
| DeepSeek V4 Pro 0813 DeepSeek | $2.22 DeepSeek (peak rate) | $1.11 | DeepSeek: $1.32 / $3.96 OpenRouter: $0.66 / $1.98 |
| Qwen3.8 Max (0902) Qwen | $3.36 Alibaba Cloud Model Studio (Qwen) (International) | $3.36 | Alibaba Cloud Model Studio (Qwen): $2.00 / $6.00 OpenRouter: $2.00 / $6.00 |
| Grok 4.7 SpaceXAI | $3.36 xAI | $3.36 | xAI: $2.00 / $6.00 OpenRouter: $2.00 / $6.00 |
| Claude Sonnet 5.5 Anthropic | $4.00 Anthropic | $4.00 | Anthropic: $2.00 / $10.00 OpenRouter: $2.00 / $10.00 |
| GPT-6 Sol OpenAI | $4.00 OpenAI | $4.00 | OpenAI: $2.00 / $10.00 OpenRouter: $2.00 / $10.00 |
| Gemini 3.1 Pro Preview | $4.32 Google Gemini API | $4.32 | Google Gemini API: $2.00 / $12.00 OpenRouter: $2.00 / $12.00 |
| Claude Opus 5.5 Anthropic | $8.00 Anthropic | $8.00 | Anthropic: $4.00 / $20.00 OpenRouter: $4.00 / $20.00 |
| Claude Fable 5.1 Anthropic | $20.00 Anthropic | $20.00 | Anthropic: $10.00 / $50.00 OpenRouter: $10.00 / $50.00 |
| GPT-6 Astra OpenAI | $20.00 OpenAI | $20.00 | OpenAI: $10.00 / $50.00 OpenRouter: $10.00 / $50.00 |
“At the model provider's own price” uses the price on the price page of the company named under the amount — usually the model's maker; for open-weight models, a hosting provider that publishes a price. “Through OpenRouter” uses OpenRouter's listed price, which follows the provider serving the model at the time; where it moved in the last 7 days, the range of the task cost is shown under the amount.
Batch variants
OpenRouter lists a separate “batch” variant of some of these models at a different price. Batch requests are not answered right away, so they are not comparable with the live prices above.
| Batch variant | Chat: 1,000 replies | Summaries: 100 documents | Coding: one working day | Prices used, per 1M tokens (input / output) |
|---|---|---|---|---|
| GPT-6 Astra (batch) | $12.50 | $3.05 | $10.00 | $5.00 / $25.00 |
| GPT-6 Sol (batch) | $2.50 | $0.61 | $2.00 | $1.00 / $5.00 |
| GPT-6 Luna (batch) | $0.13 | $0.03 | $0.10 | $0.05 / $0.25 |
| gpt-oss-120b (batch) | $0.07 | $0.02 | $0.06 | $0.0296 / $0.136 |
| Claude Fable 5.1 (batch) | $12.50 | $3.05 | $10.00 | $5.00 / $25.00 |
| Claude Opus 5.5 (batch) | $5.00 | $1.22 | $4.00 | $2.00 / $10.00 |
| Claude Sonnet 5.5 (batch) | $2.50 | $0.61 | $2.00 | $1.00 / $5.00 |
| Gemini 3.1 Pro Preview (batch) | $2.80 | $0.65 | $2.16 | $1.00 / $6.00 |
| Gemini 3.8 Flash (batch) | $0.94 | $0.23 | $0.75 | $0.375 / $1.875 |
| DeepSeek V4.1 Flash (batch) | $0.10 | $0.03 | $0.08 | $0.05 / $0.15 |
| Mistral Large 3 2512 (batch) | $0.48 | $0.13 | $0.42 | $0.25 / $0.75 |
Free variants
22 models on OpenRouter list $0 per token on the latest check. We do not put them in the tables: a free variant can be slower, capped or withdrawn.
Apodex: Apodex 1.1 Mini (free) (last listed on OpenRouter Oct 13, 2026) · Cohere: North Mini Code (free) · Dots Studio: Dots3-Note Preview (free) · Google: Gemma 4 26B A4B (free) · Google: Gemma 4 31B (free) · Google: Lyria 3 Clip Preview · Google: Lyria 3 Pro Preview · inclusionAI: Ling 3.0 Flash Sante (free) (last listed on OpenRouter Oct 8, 2026) · LiquidAI: LFM2.5-2.6B (free) · NVIDIA: Nemotron 3 Nano Omni (free) · NVIDIA: Nemotron 3 Super (free) · NVIDIA: Nemotron 3 Ultra (free) · NVIDIA: Nemotron 3.5 Content Safety (free) · NVIDIA: Nemotron 3.5 Lightning (free) · NVIDIA: Nemotron 3.5 Super VL (free) · Poolside: Laguna S 2.1 (free) · Poolside: Laguna XS 2.1 (free) · Qwen: Qwen3.8 27B (free) (last listed on OpenRouter Oct 5, 2026) · Space Bunny Alpha (last listed on OpenRouter Oct 5, 2026) · Thinking Machines: Inkling (free) · Thinking Machines: Inkling Small (free) · XingChen: Xing4.0 29B A4B (free)
Price history of these models
- GPT-6 Astra — official price: no change since
- GPT-6 Sol — official price: no change since
- GPT-6 Luna — official price: no change since
- gpt-oss-120b — official price: no change since
- Claude Fable 5.1 — official price: no change since
- Claude Opus 5.5 — official price: no change since
- Claude Sonnet 5.5 — official price: no change since
- Gemini 3.1 Pro Preview — official price: no change since
- Gemini 3.8 Flash — official price: no change since
- DeepSeek V4 Pro 0813 — official price: no change since
- DeepSeek V4.1 Flash — official price: no change since
- Grok 4.7 — official price: no change since
- Qwen3.8 Max (0902) — official price: no change since
- Qwen3.8 Flash — official price: no change since
- Mistral Large 3 2512 — official price: no change since
Official prices are compared check by check; a change counts only when a check on a later day confirms it. A model page draws the price as a chart only when the official price has changed.
How much you can send: API rate limits
Cost is one limit; the other is how many requests and tokens a new account may send per minute and how much it may spend per month. The official tiers of the Claude API, the OpenAI API and the Gemini API, with their numbers and dates, are in API rate limits and spend caps by tier.
What this comparison does not show
- Quality, speed and reliability: we do not test models and do not repeat other people's benchmarks.
- Which model to pick: the cheapest model for a task is not the best one for it.
- The models were chosen as current flagship models of the providers developers search for most; the list is fixed in our configuration and changes only by an editorial decision.
Not building on an API? Ready-made AI tools for writing, speech and images charge per month or per credit instead of per token: AI tool prices and the usage each plan includes.