Gemma 4 31B
Google Gemma 4 31B: dense open-weights model built from Gemini 3 research, with built-in thinking, multi-token prediction, and fine-tuning support.
Model Overview
Capabilities, design details, and architectural traits
Gemma 4 31B - Dense Open Model for Fine-Tuning
Gemma 4 31B is Google's dense model in the Gemma 4 family. Within the Gemma 4 lineup, it is the variant that maximizes raw quality and serves as the primary foundation for fine-tuning, contrasting with the 26B MoE variant which optimizes for inference latency. The entire Gemma 4 family is built from Gemini 3 research and released as open weights.
| Trait | Detail |
|---|---|
| Architecture | Dense 31B - not Mixture-of-Experts; all parameters active during inference, prioritizing raw output quality over speed |
| Multi-Token Prediction (MTP) | Supported; universally recommended for all tasks for this model size |
| Built-in thinking mode | Step-by-step reasoning before answering, available across the Gemma 4 family |
| Quantization checkpoints | QAT checkpoints available in compressed-tensors (-w4a16-ct) format |
Hybrid Attention Across the Gemma 4 Family
All Gemma 4 models, including the 31B, use a hybrid attention mechanism that interleaves local sliding window attention with full global attention. The final layer is always global. Global layers use unified Keys and Values with Proportional RoPE (p-RoPE) for long-context memory optimization.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | |
| Release Date | April 2, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.17 Best Input Pricing | $4 | $2 |
| Output Pricing | $0.40 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | imagetextvideo | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 13.9 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 33.2 | - | - |
| Agentic Index | 11.1 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Google
Other models by Google
Gemini 4 Argon
Gemini 3.8 Flash
Gemini 3.7 Flash
Gemini 3.6 Flash
Top AI Models
Leading alternatives by intelligence score
