Model Radar
Cost and intelligence, under the same assumptions.
Model Release Radar / Compare
Intelligence for your token budget.
The frontier marks models that no other tracked model beats on both cost and score. Set your token mix to see how caching changes the comparison. Back to the model list
Loading model prices…
Cost vs. intelligence
● Frontier · ● Other models · Lower cost is left; higher score is up
Select a point or model to inspect its prices.
| Model / provider | Score | Est. USD | Input | Output | Cache read | Cache write | Write 1h |
|---|
What this comparison means
- Estimated spend = uncached input × input rate + cache reads × read rate + cache writes × write rate + output × output rate. Read and write shares are token-weighted and cannot overlap. Write rates replace the ordinary input charge.
- These are fixed-token estimates, not measured coding-task costs. Models consume different numbers of tokens. Model detail pages keep measured DeepSWE task costs separate.
- Prices come from one named OpenRouter provider endpoint per model. The originating provider is preferred where available; otherwise we use a standard endpoint with a disclosed provider. Flex, priority, regional offers, taxes and credit-purchase fees are outside this comparison. Benchmark scores may come from a different serving configuration.
- Missing prices are unknown, never zero. A model is excluded only when this scenario needs an unknown rate, unverified write basis, unsupported context length or stale price. Google cache-write storage semantics are not yet verified.
- Capability scores are from Artificial Analysis. Each point names the highest-scoring measured effort for that metric. New releases stay unscored until benchmark data arrives.