Keep GPT-6 Sol for the work, and use Mouse-1 for the decisions around it. On a single tool-call gate, Mouse-1's median round trip was 344 ms against 1706 ms for Sol, about 5× faster, and its input costs $0.10 per 1M tokens against Sol's $2, with no output tokens to pay for.
GPT-6 Sol is OpenAI's frontier model, priced at $2 per 1M input tokens and $10 per 1M output tokens with a 1.05M-token context. Agent harnesses often send it their small judgment calls too, because it is already wired in. Mouse-1 is built only for those calls: it reads a state and a set of options and returns a probability for each one.
At a glance
| Mouse-1 | GPT-6 Sol | |
|---|---|---|
| Median verdict, one yes/no gate (our test) | 344 ms | 1706 ms |
| 90th percentile | 357 ms | 2052 ms |
| Input price per 1M tokens | $0.10 | $2 (2× above 272K) |
| Output price per 1M tokens | Free | $10 (reasoning tokens included) |
| What comes back | Calibrated probability per option | Generated text |
| Context | 16,384 state tokens | 1.05M tokens |
| Writes code, plans, explains | No | Yes |
Speed
We sent one yes/no question, "is this command safe to run without asking the user?", 20 times to each model from the same laptop, interleaving the calls so network conditions hit both alike. Sol ran at low reasoning effort with a strict JSON schema, which is how a harness would call it for a quick verdict.
Full HTTPS round trip from one client, seconds · Lower is better
| Mouse-1 | 0.34 s |
|---|---|
| GPT-6 Sol | 1.7 s |
An agent that gates every shell command makes dozens of these calls per task. At Sol's speed, those seconds add up to minutes of an agent standing still.
Cost
A decision prompt is mostly input: the task, the proposed action, the rules. Mouse-1 charges $0.10 per 1M of those tokens; Sol charges $2. Sol also bills its output and reasoning tokens at $10 per 1M, and Mouse-1 has none.
For a 2,000-token decision, that is $0.0002 with Mouse-1 and $0.004 of input alone with Sol, before any output.
How we measured
- Latency:
bench/race/latency.py, one gate, 20 interleaved calls per model from one laptop on October 20, 2026. Sol was called through OpenRouter asopenai/gpt-6-solat low reasoning effort. - Prices: OpenAI's GPT-6 Sol model page, checked October 20, 2026.
- We did not measure Sol's accuracy on our decision suites.
Sources
Checked October 20, 2026.
See also: Mouse-1 vs OpenAI Decisions · Mouse-1 vs Claude Opus as a judge · All comparisons
FAQ
Is Mouse-1 faster than GPT-6 Sol?
For a yes/no agent decision, yes. Sending the same tool-call gate to both from one client, Mouse-1's median round trip was a small fraction of GPT-6 Sol's at low reasoning effort. Sol generates an answer; Mouse-1 scores the options without generating text.
How much cheaper is Mouse-1 than GPT-6 Sol?
GPT-6 Sol costs $2 per 1M input tokens and $10 per 1M output tokens, with higher rates above 272K input tokens. Mouse-1 costs $0.10 per 1M input tokens and nothing for output: 20 times less per input token, before counting Sol's output and reasoning tokens.
Should I replace GPT-6 Sol with Mouse-1?
Not for writing code, planning or open-ended reasoning; Mouse-1 doesn't generate text. Use Mouse-1 for the narrow decisions around Sol: whether a command is safe, whether the task is done, which tools to load, which candidate is better.