| Rank | Model | Harness | Provider | Score |
|---|---|---|---|---|
| 1 | GPT-5.6 Sol | Codex | OpenAI | 43.9% |
| 2 | GPT-6 Astra | Codex | OpenAI | 42.4% |
| 3 | GPT-5.6 Sol | Pi | OpenAI | 40.9% |
| 4 | GPT-6 Astra | Pi | OpenAI | 39.4% |
| 5 | Claude Opus 5 | Pi | Anthropic | 36.4% |
| 6 | GPT-5.6 Terra | Pi | OpenAI | 34.9% |
| 7 | GPT-5.6 Terra | Codex | OpenAI | 31.8% |
| 8 | Claude Opus 5 | Claude Code | Anthropic | 30.3% |
| 9 | GPT-5.6 Luna | Pi | OpenAI | 28.8% |
| 10 | Grok 4.6 | Grok Build | SpaceXAI | 27.3% |
| 11 | DeepSeek V4.1 Flash | Pi | DeepSeek | 27.3% |
| 12 | Claude Opus 4.8 | Pi | Anthropic | 25.8% |
| 13 | Grok 4.5 | Pi | SpaceXAI | 25.8% |
| 14 | Grok 4.6 | Pi | SpaceXAI | 24.2% |
| 15 | Gemini 3.5 Flash | Pi | 24.2% | |
| 16 | GPT-5.5 | Pi | OpenAI | 22.7% |
| 17 | Claude Opus 4.8 | Claude Code | Anthropic | 21.2% |
| 18 | Claude Sonnet 5 | Pi | Anthropic | 19.7% |
| 19 | Claude Sonnet 5 | Claude Code | Anthropic | 16.7% |
| 20 | GPT-5.5 | Codex | OpenAI | 16.7% |
| 21 | Kimi K3 | Pi | Moonshot | 16.7% |
| 22 | GPT-5.6 Luna | Codex | OpenAI | 12.1% |
| 23 | Claude Opus 4.7 | Claude Code | Anthropic | 10.6% |
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.