| Rank | Model | Harness | Provider | Score |
|---|---|---|---|---|
| 1 | Claude Opus 5 | Claude Code | Anthropic | 50.3% |
| 2 | GPT-6 Astra | Codex | OpenAI | 48.6% |
| 3 | GPT-6 Astra | Pi | OpenAI | 47.1% |
| 4 | Claude Opus 4.8 | Pi | Anthropic | 44.6% |
| 5 | Grok 4.6 | Grok Build | SpaceXAI | 44.1% |
| 6 | Claude Opus 5 | Pi | Anthropic | 43.0% |
| 7 | Gemini 3.5 Flash | Pi | 42.2% | |
| 8 | Grok 4.6 | Pi | SpaceXAI | 42.0% |
| 9 | DeepSeek V4.1 Flash | Pi | DeepSeek | 38.7% |
| 10 | GPT-5.6 Sol | Pi | OpenAI | 38.3% |
| 11 | Claude Sonnet 5 | Claude Code | Anthropic | 37.2% |
| 12 | Claude Opus 4.8 | Claude Code | Anthropic | 37.0% |
| 13 | GPT-5.5 | Codex | OpenAI | 36.5% |
| 14 | Grok 4.5 | Pi | SpaceXAI | 35.6% |
| 15 | Claude Sonnet 5 | Pi | Anthropic | 33.8% |
| 16 | GPT-5.6 Terra | Pi | OpenAI | 33.5% |
| 17 | GPT-5.6 Sol | Codex | OpenAI | 32.3% |
| 18 | GPT-5.5 | Pi | OpenAI | 29.1% |
| 19 | GPT-5.6 Terra | Codex | OpenAI | 29.1% |
| 20 | GPT-5.6 Luna | Codex | OpenAI | 26.9% |
| 21 | GPT-5.6 Luna | Pi | OpenAI | 25.4% |
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.