public · reasoning
Structured JSON
Schema-constrained outputs, tool-call envelopes, and repair under stress.
Items
400
Revision
2026.05
Contamination
clean
Scorer
scorer.json.v2
- Dataset revision
- [email protected]
- Scorer
- schema-validate · scorer.json.v2
- License
- GBU public suite
- Kind
- Public reproduction suite
| Model | Score | $/task | Median s | Tools | Evidence |
|---|---|---|---|---|---|
| Claude 4.1 Opus | 96.2 | $0.267 | 17.5 | 0.2 | LIVE_MEASURED |
| o4-pro | 95.1 | $0.638 | 38.6 | 0.2 | LIVE_MEASURED |
| Claude 4.1 Sonnet | 94.8 | $0.047 | 12.0 | 0.2 | LIVE_MEASURED |
| GPT-5 | 94.2 | $0.063 | 13.3 | 0.2 | LIVE_MEASURED |
| Grok 4 Heavy | 93.1 | $0.132 | 22.9 | 0.2 | LIVE_MEASURED |
| Gemini 2.5 Pro | 92.6 | $0.025 | 8.9 | 0.2 | LIVE_MEASURED |
| Grok 4 | 91.4 | $0.038 | 10.6 | 0.2 | LIVE_MEASURED |
| Qwen3-Max | 91.1 | $0.0056 | 10.0 | 0.2 | LIVE_MEASURED |
| DeepSeek V3.1 | 90.8 | $0.0042 | 8.4 | 0.2 | REPRODUCED |
| Mistral Large 3 | 89.6 | $0.015 | 10.3 | 0.2 | LIVE_MEASURED |
| Gemini 2.5 Flash | 89.2 | $0.0011 | 3.5 | 0.2 | LIVE_MEASURED |
| GPT-5 Mini | 88.4 | $0.0031 | 5.3 | 0.2 | LIVE_MEASURED |
| Llama 4 405B | 86.4 | $0.0043 | 9.5 | 0.2 | REPRODUCED |
| Grok 3 Mini | 84.2 | $0.0016 | 4.2 | 0.2 | LIVE_MEASURED |
| Llama 4 70B | 81.2 | $0.00094 | 2.0 | 0.2 | REPRODUCED |