Alibaba flagship MoE model with 128K context. 235B parameters (22B active). Exceptional math (94/100) and reasoning (94/100). Competes with closed-source frontier models.
Epoch Capabilities Index score
Normalized ECI score (0-100)
Maximum input token capacity
Total compute used during training
Code generation, understanding, and debugging
Mathematical reasoning and problem solving
Multi-step logical reasoning
Overall reasoning and task completion ability
| Dimension | Score |
|---|---|
| Intelligence | 92.0 |
| Reasoning | 94.0 |
| Math | 94.0 |
| Code | 92.0 |
| Instruction Following | 91.0 |
| Overall | 92 |