GSO-Bench
Software optimisation: make a real codebase measurably faster without breaking it.
Agent Index weight
4%
in the Agentic coding pillar
Models tested
30
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 Supported Anthropic | 88.2% | OpenHands | — | Sep 27, 2026 | |
| 2 | GPT-6 Astra Supported OpenAI | 79.4% | OpenHands | — | Sep 27, 2026 | |
| 3 | Claude Fable 5 Supported Anthropic | 78.4% | OpenHands | — | Sep 27, 2026 | |
| 4 | GPT-5.6 Sol Supported OpenAI | 76.5% | OpenHands | — | Sep 27, 2026 | |
| 5 | Claude Opus 4.8 Supported Anthropic | 47.1% | OpenHands | — | Jul 12, 2026 | |
| 6 | Claude Opus 4.7 Supported Anthropic | 44.1% | OpenHands | — | Apr 27, 2026 | |
| 7 | Claude Opus 4.6 Supported Anthropic | 41.2% | OpenHands | — | Apr 27, 2026 | |
| 8 | GPT-5.5 Supported OpenAI | 40.2% | OpenHands | — | Apr 27, 2026 | |
| 9 | Claude Sonnet 5 Supported Anthropic | 37.3% | OpenHands | — | Jul 12, 2026 | |
| 10 | GPT-5.4 Supported OpenAI | 31.4% | OpenHands | — | Mar 10, 2026 | |
| 11 | GPT-5.2 Supported OpenAI | 27.5% | OpenHands | — | Dec 18, 2025 | |
| 12 | Claude Opus 4.5 (latest) Supported Anthropic | 26.5% | OpenHands | — | — | |
| 13 | Gemini 3.1 Pro Preview Supported Google | 22.6% | OpenHands | — | Mar 9, 2026 | |
| 14 | Gemini 3 Pro Preview Supported Google | 18.6% | OpenHands | — | Nov 28, 2025 | |
| 15 | Claude Sonnet 4.5 (latest) Supported Anthropic | 14.7% | OpenHands | — | Oct 10, 2025 | |
| 16 | GPT-5.1 Supported OpenAI | 13.7% | OpenHands | — | Dec 18, 2025 | |
| 17 | Gemini 3 Flash Preview Supported Google | 9.8% | OpenHands | — | Dec 19, 2025 | |
| 18 | o3 Supported OpenAI | 8.8% | OpenHands | — | Jun 9, 2025 | |
| 19 | Claude-Opus-4 Supported Anthropic | 6.9% | OpenHands | — | — | |
| 20 | GPT-5 Supported OpenAI | 6.9% | OpenHands | — | Aug 10, 2025 | |
| 21 | Claude Sonnet 4 Supported Anthropic | 4.9% | OpenHands | — | — | |
| 22 | Kimi K2 0905 Estimated Moonshot AI | 4.9% | OpenHands | — | Aug 26, 2025 | |
| 23 | Qwen3 Coder 480B A35B Estimated Alibaba (Qwen) | 4.9% | OpenHands | — | Sep 2, 2025 | |
| 24 | Claude-Sonnet-3.5 Estimated Anthropic | 4.6% | OpenHands | — | May 30, 2025 | |
| 25 | Gemini 2.5 Pro Supported Google | 3.9% | OpenHands | — | Oct 10, 2025 | |
| 26 | Claude 3.7 Sonnet Supported Anthropic | 3.8% | OpenHands | — | — | |
| 27 | o4-mini Estimated OpenAI | 3.6% | OpenHands | — | May 30, 2025 | |
| 28 | GLM-4.5-Air Z.ai (Zhipu) | 2.9% | OpenHands | — | Aug 26, 2025 | |
| 29 | o3-mini Estimated OpenAI | 1.3% | OpenHands | — | May 30, 2025 | |
| 30 | GPT-4o (2024-08-06) Supported OpenAI | 0.0% | OpenHands | — | May 30, 2025 |
For models without a GSO-Bench result, the index estimates one from Epoch’s Capabilities Index using a logistic fit on 28 models (residual SD 0.65 on the logit scale). Estimates are labelled and widen the model’s range.