Meta-Evaluation on Aggregated Benchmarks (AIME, ARC, GSM8K, HE, MMLU, IT, RU, BFCL)
2.4Overall Average RankGPT-5 nano
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5 nano2026.05 | 2.4 | |
| Qwen3-8BParameters=8B2026.05 | 2.6 | |
| Qwen3-4BParameters=4B2026.05 | 4.2 | |
| gpt-oss-20bParameters=20B2026.05 | 4.5 | |
| Ministral-3-8BParameters=8B2026.05 | 5 | |
| gemma-3-12b-itParameters=12B, Instruction Tuned=true2026.05 | 5.1 | |
| EngGPT2-16B-A3BParameters=16B2026.05 | 8 | |
| Llama-3.1-8BParameters=8B2026.05 | 8.2 | |
| gemma-3-4b-itParameters=4B, Instruction Tuned=true2026.05 | 9.1 | |
| Moonlight-16B-A3BParameters=16B2026.05 | 9.9 | |
| Llama-3.2-3BParameters=3B2026.05 | 10.4 | |
| LLaMAntino-3-8BParameters=8B2026.05 | 11 | |
| Velvet-14BParameters=14B2026.05 | 12.4 | |
| FastwebMIIA-7BParameters=7B2026.05 | 13.4 | |
| deepseek-moe-16bParameters=16B2026.05 | 14.5 | |
| Minerva-7BParameters=7B2026.05 | 15.2 |