GGoogle · Released Apr 2, 2026
Gemma 4 31B
Gemma 4 31B Instruct is Google DeepMind's 30.7B dense multimodal model supporting text and image input with text output. Features a 256K token context window, configurable thinking/reasoning mode, native function...
- Released
- Apr 2 2026
- Context window
- 262K 16K out
- Input / 1M tokens
- $0.20 from $0.10
- Output / 1M tokens
- $0.40 from $0.32
Where to run it
12 providers compared
Cheapest: CoreWeave at $0.10 / $0.34 per 1M input / output tokens — 80% less than the priciest option.
| Provider | Input | Output | Cache read | Context | Max output | Uptime | vs. best |
|---|---|---|---|---|---|---|---|
| CoreWeaveFP4 | $0.10 | $0.34 | $0.10 | 262K | 236K | 99.9%24h | Cheapest |
| VeniceFP4 | $0.12 | $0.36 | $0.09 | 256K | 8K | 99.7%24h | +12% |
| ChutesFP4 | $0.12 | $0.37 | $0.012 | 131K | 66K | 99.8%24h | +14% |
| DeepInfraFP8 · no tool calling | $0.16 | $0.32 | — | 262K | 236K | 95.1%24h | +25% |
| Friendli | $0.14 | $0.40 | — | 262K | 8K | 99.9%24h | +28% |
| CrusoeBF16 | $0.14 | $0.40 | $0.14 | 262K | 262K | 96%24h | +28% |
| NovitaBF16 | $0.14 | $0.40 | — | 262K | 131K | 93%24h | +28% |
| ParasailFP8 | $0.15 | $0.40 | $0.06 | 262K | 236K | 98.1%24h | +33% |
| DeepInfraFP4 | $0.14 | $0.51 | $0.075 | 262K | 16K | 99.9%24h | +43% |
| DeepInfraFP8 | $0.20 | $0.40 | — | 262K | 16K | 99.3%24h | +56% |
| DeepInfrapriority · FP8 · no tool calling | $0.30 | $0.60 | — | 262K | 236K | 98.9%24h | +134% |
| Io Net | $0.36 | $1.09 | $0.18 | 262K | 16K | 95.3%24h | +240% |
| SambaNova | $0.38 | $1.15 | — | 131K | 118K | 97.2%24h | +258% |
| ModelRunFP4 | $0.75 | $1.00 | $0.20 | 262K | 236K | 99.3%24h | +408% |
| SiliconFlowFP8 | $0.75 | $1.00 | $0.25 | 262K | 236K | 95.8%24h | +408% |
Value
Price vs. performance
Gemma 4 31B Other Google Other labs Value frontier · GPQA Diamond vs. input price per 1M tokens (log scale) · Drag to pan, pinch or Ctrl + scroll to zoom, double-click to zoom in.
Benchmarks
7 benchmark scores
| Benchmark | Score | Rank | Leader |
|---|---|---|---|
| SciCodeSoftware engineering · Research-coding problems from the natural sciences | 45.5% | #63of 93 | Claude Opus 5.566.9% |
| SimpleQA VerifiedWriting & content · Factual accuracy on 1,000 short questions | 10.4% | #66of 66 | Gemini 4 Argon76.0% |
| WebDev ArenaDesign & visual · People vote on which model builds the better website (Elo) | 1364 | #74of 88 | Claude Opus 5.51820 |
| GPQA DiamondReasoning & knowledge · PhD-level biology, chemistry and physics questions | 75.8% | #76of 92 | Gemini 4 Argon96.4% |
| CritPtReasoning & knowledge · Open-ended, research-level physics problems | 1.4% | #73of 96 | GPT-5.6 Sol32.3% |
| LMCAReasoning & knowledge · Judge the quality of conceptual arguments like experts | 39.3% | #52of 96 | Claude Sonnet 5.568.3% |
| OTIS Mock AIMEMaths · Competition-style problems harder than MATH Level 5 | 73.3% | #75of 90 | Gemini 4 Argon100.0% |
Specs
Details
- Model ID
google/gemma-4-31b-it- Developer
- Release date
- Apr 2, 2026 ·
- Context window
- 262,144 tokens
- Max output
- 16,384 tokens
- API price
- $0.20 / $0.40 per 1M input / output tokens
- Providers
- 12
- Input types
- image, text, video
- Reasoning
- Yes
- Tool calling
- Yes
- Open weights
- Yes
- Free tier
- Yes
Release date, prices and providers from OpenRouter's catalog; buying directly from a provider can cost a little differently. Uptime is the share of OpenRouter requests each provider served over the last day. Benchmark scores are the best result per benchmark from Epoch AI, licensed CC BY 4.0. Rank counts every model tracked here with a score.