Compare LLM API prices
Pick up to four models and see what each costs for your workload.
Cheapest: Qwen3.7 Flash, $31.74 / month. 99% less than Claude Opus 5.5. Highest intelligence: Claude Opus 5.5, 168.8 ECI. +24.7 ECI points vs Qwen3.7 Flash. Best value: Qwen3.7 Flash, $2.25 / ECI point. Lowest monthly cost per ECI point above 130.
Side by side
| Model |  |  | | 1 more slot free |
|---|---|---|---|---|
| Your monthly coststandard pricing | $31.74 (best in row) | $2,244 | $4,488 | |
| Per year12 months of the above | $381 (best in row) | $26,928 | $53,856 | |
| Per 1,000 requestsat your token counts | $0.106 (best in row) | $7.48 | $14.96 | |
| Input priceper 1M tokens | $0.030 (best in row) | $2.00 | $4.00 | |
| Cached inputper 1M tokens | $0.006 (best in row) | $0.100 | $0.200 | |
| Output priceper 1M tokens | $0.130 (best in row) | $10.00 | $20.00 | |
| Context windowmax tokens | 1M | 1M | 1M | |
| Intelligence (ECI)Epoch Capabilities Index, 130 = Claude 3.5 Sonnet | 144.1 | 166.2 | 168.8 (best in row) | |
| GPQA Diamondgraduate-level science, 0–100 | 79.6 | 98.0 (best in row) | 91.1 | |
| Price stabilitylast 30 days | Stable | Stable | Stable |
Shaded cell = best in row (lowest for cost and price, highest for scores, context and speed).
How to choose between models
A cheaper headline price isn’t always the cheaper model for your workload. Which rate matters most depends on the shape of your traffic, so the comparison lines up the numbers that actually drive a bill:
- Input vs output price. Which rate matters more depends on your traffic shape — see Input vs Output Tokens for why output usually costs more and how to weigh the two.
- Cached input. If you reuse a big fixed prompt or document across many calls, a model with an aggressive cached-input discount can beat a nominally cheaper model that lacks one — see Prompt Caching Explained.
- Context window. A larger window lets you pass more context in one call, but remember you pay for every token you send — a big window is a capability, not a discount.
- Monthly cost. The first row prices your own workload — input and output tokens per request, requests per day, cache share and standard or batch pricing — at each model’s rates, so you get one comparable dollar figure per model. The best value in every row is highlighted automatically.
Add up to four columns, pick any provider’s model in each, and read down the rows. When you’ve shortlisted, model your real traffic on the calculator, or weigh capability against price on the benchmarks. Full rates for every model live on the pricing page.
Frequently asked questions
Which model is cheapest overall?
There is no single answer, because it depends on your input/output ratio and cache usage. A model with cheap input but expensive output wins for short-answer workloads and loses for long-generation ones. The monthly cost row prices each model for the workload you enter, so it is a like-for-like figure; confirm it with the calculator using your real traffic.
What does the monthly cost row assume?
It uses the workload shown above the table: input and output tokens per request, requests per day, the share of input served from the prompt cache, and standard or batch pricing. Requests are multiplied over a 30-day month and priced at each model's own input, cached-input and output rates. It is a shared yardstick for comparison, not a prediction of your bill.
How many models can I compare at once?
Up to four. Each column has its own searchable model picker, and the table scrolls sideways on small screens. The best value in each attribute row is flagged automatically.
Does a bigger context window cost more?
The window itself is a capacity limit, not a charge. But you pay per token for everything you actually send, so filling a large window with context increases the input cost of each call.
Can I share or bookmark a comparison?
Yes. The chosen models and your workload are saved in the page address, so copying the link reopens the same comparison. Your workload is also remembered in this browser and shared with the other pages on the site.