| Rank | Model | Harness | Provider | Score |
|---|---|---|---|---|
| 1 | GPT-5.5 | Pi | OpenAI | 45.0% |
| 2 | GPT-6 Astra | Codex | OpenAI | 44.6% |
| 3 | GPT-5.6 Sol | Codex | OpenAI | 42.1% |
| 4 | DeepSeek V4.1 Flash | Pi | DeepSeek | 42.1% |
| 5 | GPT-6 Astra | Pi | OpenAI | 41.8% |
| 6 | GPT-5.5 | Codex | OpenAI | 39.9% |
| 7 | GPT-5.6 Sol | Pi | OpenAI | 39.9% |
| 8 | GPT-5.6 Terra | Pi | OpenAI | 39.9% |
| 9 | GPT-5.6 Terra | Codex | OpenAI | 39.3% |
| 10 | Claude Opus 4.8 | Pi | Anthropic | 39.0% |
| 11 | Claude Sonnet 5 | Claude Code | Anthropic | 37.7% |
| 12 | Claude Opus 4.8 | Claude Code | Anthropic | 37.4% |
| 13 | Grok 4.6 | Grok Build | SpaceXAI | 37.4% |
| 14 | GPT-5.6 Luna | Pi | OpenAI | 36.5% |
| 15 | Claude Opus 5 | Pi | Anthropic | 35.2% |
| 16 | Claude Sonnet 5 | Pi | Anthropic | 35.2% |
| 17 | Grok 4.6 | Pi | SpaceXAI | 34.3% |
| 18 | Claude Opus 4.7 | Claude Code | Anthropic | 31.4% |
| 19 | GPT-5.6 Luna | Codex | OpenAI | 31.4% |
| 20 | Claude Opus 4.7 | Pi | Anthropic | 31.1% |
| 21 | Claude Opus 5 | Claude Code | Anthropic | 31.1% |
| 22 | Grok 4.5 | Pi | SpaceXAI | 28.9% |
| 23 | Gemini 3.5 Flash | Pi | 28.9% | |
| 24 | Kimi K3 | Pi | Moonshot | 26.7% |
Methodology and provenance
Scores use a 0–100 percentage scale. See the benchmark repository, hosted paper PDF, and original publication for task construction, grading, aggregation, and uncertainty details.