Gemma 4 31B
Google DeepMind · released
Google's strongest open-weights model: a 31-billion-parameter multimodal model with an optional thinking mode that fits on a single 80GB GPU, or on consumer GPUs when quantised. Permissively licensed and meant for self-hosting, fine-tuning and local agents.
Best for: Running locally · Fine-tuning · On-device agents
- Intelligence Index
- 14.7
- #34 of 38 · at Reasoning effort
- Price per 1M tokens
- $0.09 in $0.34 out
- Blended $0.15 at 3:1
- Context window
- 262K
- Output speed
- 35 tokens/s
- Median, measured by Artificial Analysis
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall26 · #32 of 38
Based on 2 of 3 benchmarks
- Coding13 · #31 of 39
Based on 3 of 5 benchmarks
- Agents & knowledge work9 · #36 of 40
Based on 1 of 6 benchmarks
- Reasoning & science16 · #33 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability22 · #36 of 39
Based on 2 of 2 benchmarks
- Vision56 · #23 of 31
Based on 2 of 2 benchmarks
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#34 of 38· Reasoning effort· Artificial Analysis
- Vals IndexNot tested
- Arena · Text1453 ±7
#21 of 33· 6,133 votes· Arena
Coding
#35 of 38· Reasoning effort· Artificial Analysis
- SciCode45.5%
#31 of 38· Reasoning effort· Artificial Analysis
- Arena · WebDev1365 ±6
#32 of 34· 13,592 votes· Arena
- Vibe Code BenchNot tested
- Code MigrationNot tested
Agents & knowledge work
- GDPval-AA6.1
#35 of 39· Reasoning effort· Artificial Analysis
- Arena · AgentNot tested
- Finance Agent v2Not tested
- Excel Modeling BenchmarkNot tested
- Legal Research BenchNot tested
- Tax Agent BenchNot tested
Reasoning & science
- Humanity's Last Exam23.6%
#32 of 39· Reasoning effort· Artificial Analysis
- CritPt1.4%
#33 of 39· Reasoning effort· Artificial Analysis
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA Diamondsaturating85.7%
#20 of 27· Reasoning effort· Artificial Analysis
Knowledge & reliability
#37 of 39· Reasoning effort· Artificial Analysis
- AA-LCR69.7%
#34 of 39· Reasoning effort· Artificial Analysis
Vision
- MMMU-Pro73.4%
#22 of 27· Reasoning effort· Artificial Analysis
- Arena · Vision1262 ±6
#12 of 19· 39,111 votes· Arena
Reported by Google DeepMind
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- GPQA Diamond84.3%
- MMMU-Pro76.9%
Specifications
- Released
- Apr 2, 2026
- Availability
- Generally available
- Context window
- 262,144 tokens
- Max output
- —
- Accepts
- Text, Images, Video
- Knowledge cutoff
- Jan 2025
- Weights
- Open
- Licence
- Apache 2.0
- Parameters
- 30.7B
- Reasoning control
- Optional thinking mode
- API model ID
gemma-4-31b-it
Pricing
- Input / 1M tokens
- $0.09
- Cached input / 1M
- $0.05
- Output / 1M tokens
- $0.34
- Blended (3:1)
- $0.15
Typical price at third-party hosts. The lab does not sell paid API access to this model.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.