Decision tool
Which model?
Three choices, five picks — ranked from the same verified data as the rest of the terminal, with the scoring rule in plain sight. No quiz, no black box.
300 candidate models · verified Jul 27, 2026 · methodology
For
Optimize
Limits
| # | Pick | Cap | $/1M | tok/s | |
|---|---|---|---|---|---|
| 1 | Kimi K3Kimiopen | 99.2 | $6 | 33 | quality leader |
| 2 | Claude Opus 5Anthropic | 97.7 | $10 | 44 | −1.5 pts vs best |
| 3 | Gemini 3.1 Pro PreviewGoogle | 97.5 | $4.5 | 132 | −1.7 pts vs best |
| 4 | GPT-5.5OpenAI | 97.0 | $11.25 | — | −2.2 pts vs best |
| 5 | GPT-5.6 Luna (max)OpenAI | 96.6 | $2.25 | 171 | −2.6 pts vs best |
Compare top 3 →
Quality = cross-benchmark capability. Budget/Fastest pick among models within 15 pts of the best. Balanced = quality − 8·log₁₀(price) + 3·log₁₀(speed). Same data as everywhere on Modelyst — sources on every model page.
Want the full picture instead? The frontier shows every model on the capability-price plane, and the cost calculator prices your exact workload.