| Rank | Model | Harness | Provider | Score |
|---|---|---|---|---|
| 1 | Grok 4.6 | Grok Build | SpaceXAI | 77.4% |
| 2 | DeepSeek V4.1 Flash | Pi | DeepSeek | 75.4% |
| 3 | GPT-5.6 Sol | Codex | OpenAI | 74.5% |
| 4 | GPT-5.6 Sol | Pi | OpenAI | 73.0% |
| 5 | Claude Opus 5 | Claude Code | Anthropic | 71.0% |
| 6 | Grok 4.5 | Pi | SpaceXAI | 71.0% |
| 7 | GPT-5.6 Terra | Pi | OpenAI | 70.4% |
| 8 | GPT-6 Astra | Codex | OpenAI | 70.1% |
| 9 | GPT-6 Astra | Pi | OpenAI | 70.1% |
| 10 | Grok 4.6 | Pi | SpaceXAI | 69.9% |
| 11 | GPT-5.6 Terra | Codex | OpenAI | 69.6% |
| 12 | Claude Sonnet 5 | Pi | Anthropic | 69.3% |
| 13 | Claude Opus 4.8 | Claude Code | Anthropic | 69.0% |
| 14 | Gemini 3.5 Flash | Pi | 67.0% | |
| 15 | Claude Opus 5 | Pi | Anthropic | 66.7% |
| 16 | Claude Opus 4.8 | Pi | Anthropic | 66.4% |
| 17 | GPT-5.5 | Pi | OpenAI | 65.2% |
| 18 | GPT-5.5 | Codex | OpenAI | 64.9% |
| 19 | Claude Sonnet 5 | Claude Code | Anthropic | 64.3% |
| 20 | Claude Opus 4.7 | Claude Code | Anthropic | 64.1% |
| 21 | Claude Opus 4.7 | Pi | Anthropic | 63.8% |
| 22 | Kimi K3 | Pi | Moonshot | 61.5% |
| 23 | GPT-5.6 Luna | Codex | OpenAI | 60.6% |
| 24 | GPT-5.6 Luna | Pi | OpenAI | 57.1% |
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.