Benchgen

MGSM — Results

RankModelScore
1deepseek-v4-1-flash80.2
2llama-4-maverick0.923
3o3-mini0.92
4claude-3-5-sonnet0.916
5llama-3-3-70b-instruct0.911
6llama-4-scout0.906
7gpt-4o0.905
8o10.893
9gpt-4-turbo0.885
10gemini-1-5-pro0.875
11gpt-4o-mini0.87
12llama-3-2-90b-instruct0.869
13claude-3-5-haiku0.856
14qwen3-235b-a22b0.835
15gemini-1-5-flash0.826
16phi-40.806
17phi-4-mini0.639
18phi-3-5-moe-instruct0.587
19llama-3-2-3b-instruct0.582
20phi-3-5-mini-instruct0.479

MGSM — Phases

1 phaseActive

Phase 1·

MGSM

Aug 3, 2026
Current