MERA Multi - UniScienceVQA: leaderboard
MERA Multi task UniScienceVQA: a private-test Russian image benchmark of university-level science questions posed with images, scored by MERA's task metric (Exact Match or judge score). MERA's 0-1 result scaled to 0-100; higher is better. From the MERA Multi leaderboard (AI Alliance Russia); only submissions MERA shows publicly for the modality are included.
Metric: Score (%). Source: mera.a-ai.ru. Saturation forecast: Around March 2027. 60 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 35B A3B | 33.1 | #238 |
| 2 | Gemma 4 31B (IT) | 31.3 | #178 |
| 3 | Qwen 3.5 27B | 31.1 | #192 |
| 4 | Qwen 3.6 27B | 30.5 | #182 |
| 5 | Qwen 3.5 35B A3B | 30.2 | #268 |
| 6 | Qwen 3.5 122B A10B | 28.9 | #184 |
| 7 | Qwen 2 VL 72B Instruct | 25.5 | #551 |
| 8 | GPT-4.1 | 25.3 | #258 |
| 9 | Gemma 4 26B A4B (IT) | 23.5 | #293 |
| 10 | Qwen3 Omni 30B A3B Instruct | 22.4 | #401 |
| 11 | Qwen 2.5 VL 72B Instruct | 18 | #383 |
| 12 | Gemma 3 27B (IT) | 16 | #528 |
| 13 | Qwen 3 VL 8B Instruct | 16 | #419 |
| 14 | Qwen 2 VL 7B Instruct | 13.9 | #821 |
| 15 | Qwen 3 VL 4B Instruct | 13.2 | #510 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mera-multi-unisciencevqa · How It Works · Data refreshed daily, snapshot 2026-11-05.