EuroEval Romanian Summarization - Sumo RO: leaderboard
EuroEval's Romanian summarization task on the Sumo RO dataset — models write abstractive summaries of Romanian news articles, scored with automatic text-overlap metrics, from the EuroEval European language model benchmark suite.
Metric: chrF3++ (%). Source: euroeval.com. Saturation forecast: Around 2034. 150 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 38.27 | #248 |
| 2 | Bielik-11B-v2.3-Instruct | 37.99 | #693 |
| 3 | Gemini 3 Pro (Preview) | 37.55 | #66 |
| 4 | Apertus-70B-Instruct-2509 | 37.51 | #820 |
| 5 | EuroLLM-9B-Instruct | 37.49 | #1054 |
| 6 | Mistral Small 3.1 | 37.27 | #610 |
| 7 | Qwen 3 30B A3B 2507 Instruct | 37.22 | #488 |
| 8 | GPT-5.4 Mini (2026-03-17) (High) | 37.14 | #169 (GPT-5.4 Mini (2026-03-17)) |
| 9 | Gemini 3 Flash (Preview) (Non-reasoning) | 37.11 | #82 (Gemini 3 Flash (Preview)) |
| 10 | Qwen 2.5 72B Instruct | 36.99 | #454 |
| 11 | Qwen 3 32B (Non-reasoning) | 36.98 | #443 (Qwen 3 32B) |
| 12 | Mistral Small 3.2 | 36.94 | #605 |
| 13 | Qwen 3 14B (Non-reasoning) | 36.82 | #537 (Qwen 3 14B) |
| 14 | Gemini 2.5 Flash Lite (Non-reasoning) | 36.81 | #422 (Gemini 2.5 Flash Lite) |
| 15 | EuroLLM-22B-Instruct-2512 | 36.81 | #701 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=euroeval-romanian-summarization-sumo-ro · How It Works · Data refreshed daily, snapshot 2026-10-20.