MERA Code: leaderboard
The total score of MERA Code, the Russian code-evaluation arm of MERA: eleven tasks spanning generation, review, test writing and repository-scale work, aggregated by the leaderboard’s own weighting.
Metric: Total score. Source: mera.a-ai.ru. Saturation forecast: Around July 2027. 66 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 Codex Max | 0.54 | #100 |
| 2 | GLM-5.1 | 0.53 | #97 |
| 3 | DeepSeek V4 Pro | 0.51 | #102 |
| 4 | GPT-5.4 | 0.5 | #79 |
| 5 | Claude Opus 4.6 | 0.48 | #63 |
| 6 | DeepSeek V4 Flash | 0.47 | #137 |
| 7 | Qwen 3 Next 80B A3B (Thinking) | 0.47 | #320 (Qwen 3 Next 80B A3B) |
| 8 | Qwen 3 30B A3B 2507 (Thinking) | 0.47 | #376 (Qwen 3 30B A3B 2507) |
| 9 | Qwen 3 32B | 0.46 | #443 |
| 10 | Qwen 3 4B 2507 (Thinking) | 0.45 | #538 (Qwen 3 4B 2507) |
| 11 | MiniMax-M2.7 | 0.43 | #258 |
| 12 | Qwen 3 14B | 0.41 | #537 |
| 13 | Qwen 3 4B | 0.39 | #829 |
| 14 | GPT-4.1 | 0.38 | #248 |
| 15 | Qwen 3 8B | 0.37 | #674 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mera-code · How It Works · Data refreshed daily, snapshot 2026-10-20.