Skip to main content
Correctness review uses detection modes. For Check Run Agents, you choose a model in the definition’s front matter. The default is claude-opus-5-5.

​
Choose a model

For a model that uses reasoning instead:
Use the rate table to compare costs and MacroscopeBench to compare measured review quality, cost, and duration at specific reasoning levels. An unavailable or retired model ID skips the check with a “model is not available on Macroscope” message. It does not fall back to another model.

​
Reasoning and effort

Both fields default to low. An unsupported reasoning level warns and falls back to low. Some models map several levels to the same behavior; see the notes below.

​
Available models

muse-wonder-1-0 runs on Meta’s hosted API. Open-source models use Fireworks AI hosted inference. Custom endpoints and self-hosted models are not supported; provider queuing can affect latency.

​
Pricing

Raw USD per 1 million tokens. Each product adds its own billing rules: Check Run Agents, Macroscope pricing, and Murmur. Input, output, cache-write, and cached-input counts do not overlap. Output includes reasoning tokens. A dash means no separate cache-write charge. Model IDs below match Macroscope configuration, which may differ from provider spelling.

​
Long-context rates

Crossing a threshold reprices the whole request. Prompt size includes uncached input, cached input, and cache writes.
  • gpt-5-4, gpt-5-5, gpt-5-6-sol, gpt-5-6-terra, gpt-5-6-luna, gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-6-1-sol, gpt-6-1-terra: above 272,000 prompt tokens: 2x input, 1.5x output, and 2x on both cache columns where they are charged.
  • grok-4-6: at 200,000 prompt tokens or more: 2x on every token, output included.
Other listed models use the same rate at all prompt sizes.