Google DeepMindopen weightsCurrentReasoning
Gemma 4 31B
Gemma moved to Apache-2.0 with this generation.
Apr 2, 2026
—
—
256K
Benchmark scores
| Benchmark | Score | Δ vs prior | Rank (current) | Configuration | Source |
|---|---|---|---|---|---|
SWE-bench Verified | 52% | #8 of 8 | per Qwen's comparison table · leaderboard | morphllm.com ↗ | |
GPQA Diamond | 84.3% | #17 of 21 | lab-reported, thinking · lab-reported | huggingface.co ↗ | |
Humanity's Last Exam | 19.5% | #28 of 31 | no tools; Google reports 26.5 with tools · lab-reported | huggingface.co ↗ | |
τ²-bench (Telecom) | 59.9% | #8 of 9 | leaderboard | benchlm.ai ↗ | |
MMLU-Pro | 85.2% | #9 of 12 | leaderboard | benchlm.ai ↗ | |
Artificial Analysis Intelligence Index | 29.4 | #14 of 16 | leaderboard | benchlm.ai ↗ |