LLM API pricing comparison
Every model we track, sorted from cheapest to most expensive by combined input and output price. Prices are USD per 1M tokens at list rates, verified 2026-10-20.
| Model | Vendor | Input | Output | Cached in | Context | 1K requests* |
|---|---|---|---|---|---|---|
| GPT-5 nano | OpenAI | $0.05 | $0.40 | $0.005 | 400K | $0.25 |
| DeepSeek V3.2 (chat) | DeepSeek | $0.28 | $0.42 | $0.03 | 128K | $0.49 |
| GPT-4o mini | OpenAI | $0.15 | $0.60 | $0.07 | 128K | $0.45 |
| Llama 4 Maverick | Meta | $0.27 | $0.85 | — | 1M | $0.69 |
| GPT-5 mini | OpenAI | $0.25 | $2.00 | $0.03 | 400K | $1.25 |
| DeepSeek R1 (reasoner) | DeepSeek | $0.55 | $2.19 | $0.14 | 128K | $1.65 |
| Gemini 2.5 Flash | $0.30 | $2.50 | $0.07 | 1M | $1.55 | |
| Kimi K2 | Moonshot | $0.60 | $2.50 | — | 262K | $1.85 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | $0.10 | 200K | $3.50 |
| Qwen3 Max | Alibaba | $1.20 | $6.00 | — | 262K | $4.20 |
| Mistral Large 2 | Mistral | $2.00 | $6.00 | — | 128K | $5.00 |
| GPT-5 | OpenAI | $1.25 | $10.00 | $0.13 | 400K | $6.25 |
| Gemini 2.5 Pro | $1.25 | $10.00 | $0.31 | 1M | $6.25 | |
| GPT-4o | OpenAI | $2.50 | $10.00 | $1.25 | 128K | $7.50 |
| Claude Sonnet 4.5 | Anthropic | $3.00 | $15.00 | $0.30 | 200K | $10.50 |
| Grok 4 | xAI | $3.00 | $15.00 | — | 256K | $10.50 |
| Claude Opus 4.1 | Anthropic | $15.00 | $75.00 | $1.50 | 200K | $52.50 |
* 1K requests assumes a 1,000-token prompt and 500-token response each. Actual spend varies with prompt size, caching and batch discounts.
How to read these numbers
Output is the multiplier. Across every vendor, output tokens cost 2x to 8x more than input. A model that looks cheap on input can be expensive in production if your application generates long answers. Sort by the 1K requests column to see the blended cost for a typical chat workload.
Caching changes the math. If a large system prompt is reused across requests, cached input can cut that portion of your bill by 50-90%. Workloads above roughly 5,000 requests per day almost always benefit from enabling it.
Context length is not free. A 1M-token window is impressive, but a single call that actually fills it costs hundreds of times a normal request. Use the calculator on your real documents before assuming a long-context model is affordable.
Have a specific prompt in mind?
The calculator counts exact tokens for your text and prices it across every model in this table.
Open the token calculator →