LLM API pricing comparison
55 current models ranked by what the same workload costs. Change the workload and the whole table re-ranks on the server.
At 1,000 requests per day with 3,000 input and 600 output tokens each, the cheapest current model is GPT-5 nano at $11.70 per month and the most expensive is GPT-5 pro at $3,510.00, a 300 times the price of the cheapest.
Reading the table
- Output tokens dominate for generation-heavy workloads. Set output tokens to your real average, not the maximum.
- Input price alone can mislead. On this dataset GPT-5.4 lists $2.50 input and $15.00 output per 1M tokens, and Claude Sonnet 5 lists $2.00 and $10.00.
- Caching changes the ranking for repeated prefixes. The calculator includes cache fields; this table does not.
- The price of a cheaper model is only a saving if quality holds. SpendLint reports cost only.
Methodology
- Scope: current models on the official pricing pages of Anthropic, OpenAI, Google and Azure OpenAI. The dataset holds 58 models and 216 price rows; superseded rows stay in it with an end date, and retired models are left out of the table.
- Unit: US dollars per one million tokens at list price. Negotiated discounts, commitment tiers, regional pricing, taxes and free credits are not included.
- Per request: input tokens times the input price plus output tokens times the output price, with long-context tiers applied where a provider publishes them, and the batch multiplier when you select it.
- Per month: the per-request cost times requests per day times 30 days.
- Verification: each row records the date it was last checked against the official source, and the Source column links that source. The newest check in this dataset is .
Cite or reuse this data
You can reuse the full dataset under CC BY 4.0 with attribution to SpendLint and to the official source named on each row. Download it as JSON or CSV. Each row in the table has a permalink, for example https://spendlint.dev/compare/llm-api-pricing#anthropic-claude-sonnet-5, and a query string such as ?rpd=5000&in=2000&out=400 reproduces a ranked workload.
SpendLint. "LLM API pricing comparison." Prices verified 2026-09-05. https://spendlint.dev/compare/llm-api-pricing
To price your own workload in detail, including cache reads and writes, use the LLM cost calculator. For Claude Code seat and usage cost, see Claude Code cost.
Sources and verification
- Anthropic: https://platform.claude.com/docs/en/about-claude/pricing
- OpenAI: https://developers.openai.com/api/docs/pricing
- Google (Gemini API): https://ai.google.dev/gemini-api/docs/pricing
- Azure OpenAI (Foundry Models): https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/