Thinkingbox: leaderboard
Thinkingbox executable task pass@1, micro-averaged over 20 trials for each of 507 business-workflow tasks. Source-trained GRPO checkpoints remain distinct from their parent models. Source reports medium reasoning requests for API agents but says Claude request settings and returned IDs are unverified.
Metric: Pass@1 (%, 20 trials per task). Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 66.5 | #12 |
| 2 | GPT-5.4 (Medium) | 65.36 | #79 (GPT-5.4) |
| 3 | GPT-5.6 Sol (Medium) | 61.91 | #19 (GPT-5.6 Sol) |
| 4 | Claude Sonnet 4.6 | 59.19 | #90 |
| 5 | GPT-6 Astra (Medium) | 58.31 | #6 (GPT-6 Astra) |
| 6 | Qwen 3.8 27B | 51.7 | #105 |
| 7 | Claude Opus 4.6 | 32.09 | #63 |
| 8 | Grok 4.3 (Medium) | 14.38 | #188 (Grok 4.3) |
| 9 | Qwen 3.5 9B | 5.65 | #379 |
| 10 | Mistral Large 3 | 4.66 | #401 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=thinkingbox · How It Works · Data refreshed daily, snapshot 2026-10-20.