Revelo Code Index - SWE-Lancer
Code · 2026-07-29
The Revelo Code Index's SWE-Lancer task set: 20 private freelance software jobs posted on Freelancer, Upwork, RedditHire and 99Lancers, carried over with the client's original specification and closing price in the format of OpenAI's SWE-Lancer, though none are SWE-Lancer's own tasks. A trial passes only when every required held-out executable test succeeds against the finished application, and the score is the mean over tasks of three trials; prices are not used in scoring.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 51.7 |
| Gemini 3.8 Flash | 51.7 |
| Grok 4.5 | 46.7 |
| Opus 4.8 | 45.0 |
| Claude Opus 5 | 45.0 |
| GPT-5.6 Sol | 43.3 |
| Kimi K3 | 43.3 |
| Gemini 3.6 Flash | 41.7 |
| GLM-5.2 | 40.0 |
| GPT-5.6 Terra | 40.0 |
| GPT-6 Astra | 40.0 |
| GPT-5.6 Luna | 38.3 |
| DeepSeek V4.1 Flash | 35.0 |
| DeepSeek-V4-Flash-0731 | 35.0 |
| Qwen3.8-Max | 28.3 |
| DeepSeek-V4-Pro | 26.7 |
| MiniMax M3 | 21.7 |
| Hy3 | 16.7 |
| MiMo-V2.5-Pro | 6.7 |