Gemma 4 31B
Google · Released 31 Mar 2026 · gemma-4-31b-it
–
Input price / 1M tokens
–
Output price / 1M tokens
262K
Context window
#54 of 71
35
Output tokens / second
#59 of 61
80%
LiveCodeBench
#2 of 7
84.3%
GPQA Diamond
#20 of 27
76.9%
τ²-bench
#5 of 6
19.5%
Humanity’s Last Exam (no tools)
#17 of 23
Summary
Gemma 4 31B is an open-weight model from Google, released on 31 Mar 2026.
Its 262K-token context window ranks #54 of 71.
Measured output speed is 35 tokens per second, #59 of 61.
Largest dense Gemma 4 open model (30.7B parameters), with a thinking mode, function calling and 256K context. Video is handled as frame sequences.
Benchmarks · 10 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| MMLU Proinstruction-tuned | 85.2% | – | Gemma 4 model cardSelf-reported | |
| AIME 2026no tools | 89.2% | – | Gemma 4 model cardSelf-reported | |
| LiveCodeBench | 80% | #2 of 7 | Gemma 4 model cardSelf-reported | |
| Codeforces | 2,150 | Elo-style | – | Gemma 4 model cardSelf-reported |
| GPQA Diamond | 84.3% | #20 of 27 | Gemma 4 model cardSelf-reported | |
| τ²-benchaverage over 3 | 76.9% | #5 of 6 | Gemma 4 model cardSelf-reported | |
| Humanity’s Last Exam (no tools)no tools (26.5% with search) | 19.5% | #17 of 23 | Gemma 4 model cardSelf-reported | |
| BigBench Extra Hard | 74.4% | – | Gemma 4 model cardSelf-reported | |
| MMMU-Pro | 76.9% | #4 of 12 | Gemma 4 model cardSelf-reported | |
| MRCR v2 (8-needle)128k average | 66.4% | #8 of 13 | Gemma 4 model cardSelf-reported |
Self-reported means the lab ran the test itself. A rank appears only where several labs report the same version of a benchmark.
Similar models
Models with the most benchmarks in common with Gemma 4 31B, and the closest scores.
LiveCodeBench
#2 of 7Fresh competitive programming problems
GPQA Diamond
#20 of 27Graduate-level science questions
- Gemma 4 31B84.3
- Gemma 4 26B A4B82.3
- Gemini 3.1 Flash-Lite86.9
- Gemini 3 Flash90.4
- Gemini 3.1 Pro94.3
- Gemini 2.5 Flash-Lite66.7
τ²-bench
#5 of 6Tool use with a simulated customer
- Gemma 4 31B76.9
- Gemma 4 26B A4B68.2
- Gemini 3 Flash90.2
- Gemini 3.1 Pro90.8
Humanity’s Last Exam (no tools)
#17 of 23Expert questions, no search or code
Compare Gemma 4 31B side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- license, context, modalities, parameters, knowledge cutoff, benchmarks: ai.google.dev/gemma/docs/core/model_card_4, checked 2 Oct 2026
- release date: ai.google.dev/gemma/docs/releases, checked 2 Oct 2026
- Gemini API availability and id: ai.google.dev/gemini-api/docs/changelog, checked 2 Oct 2026
- Gemini API price (free only): ai.google.dev/gemini-api/docs/pricing, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/gemma-4-31b, checked 2 Oct 2026
Building on Gemma 4 31B?
Get the architecture right before the bill arrives.
We size caching, routing and fallbacks for your workload on a free call, and tell you where a cheaper model is good enough.