Google DeepMindopen-weight
Released: 2026-04-02Artificial AnalysisGemma 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...
Intelligence Index
4.4/ 100
Artificial Analysis Intelligence Index (independent composite)Throughput Speed
32tok/sec
Output streaming throughputTime to First Answer
52s
Time to first answer token, including any thinkingCost to Run the Index
—USD
Artificial Analysis cost of the full Index runMeasured results · Artificial Analysis
Independent evaluations of Gemma 4 31B (Reasoning configuration). Values are reported as published; nothing here is estimated or derived. Snapshot 2026-11-15.
- Intelligence Index
- 4.4
- Cost to run the Index
- —
- Time per task
- —
- Omniscience index
- -48.5
- Reasoning model
- Yes
- Cached-input price
- —
Reasoning & knowledge
- AA-Omniscience Accuracy19.9%
- CritPt1.4%
- CritPt v2 (Hard)4.0%
- GPQA Diamond85.7%
- HLE-Diamond7.2%
- Humanity's Last Exam23.6%
- IFBench75.6%
Coding & terminal agents
- FrontierCode-AA 2.00.0%
- SciCode45.5%
- Terminal-Bench 2.143.5%
- Terminal-Bench 4.00.0%
- Terminal-Bench Hard36.4%
- Terminal-Bench-Science 0.10.0%
Agentic work
- AA-Briefcase v1.1 (Elo)369 Elo
- AutomationBench-AA4.9%
- EnterpriseOps-Gym-AA28.3%
- GDPval-AA (normalized Elo)627 Elo
- ITBench-AA37.3%
- τ²-Bench Telecom59.9%
- τ³-Banking14.8%
Long context & documents
Multimodal & science
- MMMU-Pro73.4%
Health & safety
Reasoning-effort variants evaluated
| Variant | Intelligence Index | Cost to run Index | Output speed (tok/s) |
|---|---|---|---|
| Reasoning | 4.4 | — | 31.6 |
| Non-reasoning | — | — | 53.1 |
Headline Evaluation Breakdown
Six independently measured evaluations (Artificial Analysis)Reasoning · CritPt v2 (Hard)4.0%
Long context · AA-LCR69.7%
Knowledge · Humanity's Last Exam23.6%
Instruction following · IFBench75.6%
Sourced Benchmark Evaluations (22)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| AA-LCR (Long Context Reasoning) | Active | 69.7% | 93 | 2026-11-15 | independent | Source ↗ |
| ITBench-AA | Active | 37.3% | 93 | 2026-11-15 | independent | Source ↗ |
| EnterpriseOps-Gym-AA | Active | 28.3% | 93 | 2026-11-15 | independent | Source ↗ |
| AutomationBench-AA | Active | 4.9% | 93 | 2026-11-15 | independent | Source ↗ |
| GDP.pdf | Active | 6% | 93 | 2026-11-15 | independent | Source ↗ |
| FrontierCode-AA 2.0 | Active | 0% | 93 | 2026-11-15 | independent | Source ↗ |
| AA-Omniscience (Non-Hallucination) | Active | 14.5% | 93 | 2026-11-15 | independent | Source ↗ |
| AA-Omniscience (Accuracy) | Active | 19.9% | 93 | 2026-11-15 | independent | Source ↗ |
| HLE-Diamond | Active | 7.2% | 93 | 2026-11-15 | independent | Source ↗ |
| CritPt v2 (Hard) | Active | 4% | 93 | 2026-11-15 | independent | Source ↗ |
| CritPt | Active | 1.4% | 93 | 2026-11-15 | independent | Source ↗ |
| GPQA | Nearing Saturation | 85.7% | 78 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench-Science 0.1 | Active | 0% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench 4.0 | Active | 0% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench 2.1 | Active | 43.4% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench Hard | Active | 36.4% | 93 | 2026-11-15 | independent | Source ↗ |
| τ³-Banking | Active | 14.8% | 93 | 2026-11-15 | independent | Source ↗ |
| τ²-Bench Telecom | Active | 59.9% | 93 | 2026-11-15 | independent | Source ↗ |
| SciCode | Active | 45.5% | 93 | 2026-11-15 | independent | Source ↗ |
| IFBench | Active | 75.6% | 93 | 2026-11-15 | independent | Source ↗ |
| MMMU-Pro | Active | 73.4% | 93 | 2026-11-15 | independent | Source ↗ |
| Humanity's Last Exam | Active | 23.6% | 93 | 2026-11-15 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run google-gemma-4-31b
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "google-gemma-4-31b",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Gemma 4 31B represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.