Side by side
See the difference.
Scores, capabilities and costs for the models you’re considering.
Scroll to compare models →
| 2 models | googleGoogle: Gemma 4 31BRemove × | NVIDIANemotron 3.5 Super VLRemove × |
|---|---|---|
| What’s different | 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… | 120B/12B-active Mamba-2 hybrid with a vision encoder; native speculative decoding and switchable thinking. |
| Date | Mar 11, 2026Listed | Oct 15, 2026Released |
API prices USD · provider configuration shown below | ||
| Listed rate | $0.10 in / $0.34 out/ 1M tokensCoreWeave · via OpenRouter · all routes ↗coreweave/fp4 · quantization=fp4 | See provider quotes |
| Provider endpoints | 20 endpoints → | 0 endpoints → |
Benchmarks Each result retains its evaluation configuration | ||
| AA Intelligence ↗4.3.2 | Not reported | 26.49 ↗Effort not specifiedIndependent result |
| AA-Briefcase ↗1.1 | Not reported | 998.76 ↗Effort not specifiedIndependent result |
| GDPval-AA ↗2.1 | Not reported | 1,162.41 ↗Effort not specifiedIndependent result |
| AutomationBench-AA ↗AA Index 4.3.2 collection | Not reported | 14.82% ↗Effort not specifiedIndependent result |
| Terminal-Bench 4.0 · AA ↗ | Not reported | 0% ↗Effort not specifiedHarness unreportedIndependent result |
| SciCode ↗AA Index 4.3.2 collection | Not reported | 47.8% ↗AA · Under reviewEffort not specifiedIndependent result |
| Humanity’s Last Exam ↗AA Index 4.3.2 collection | Not reported | 30.72% ↗Effort not specifiedIndependent result |
| GDP.pdf ↗AA Index 4.3.2 collection | Not reported | 8% ↗Effort not specifiedIndependent result |
| CritPt ↗AA Index 4.3.2 collection | Not reported | 10.57% ↗AA · Under reviewEffort not specifiedIndependent result |
| AA-Omniscience ↗AA Index 4.3.2 collection | Not reported | -4.22 ↗Effort not specifiedIndependent result |
| AA-LCR ↗1.1 | Not reported | 69% ↗Effort not specifiedIndependent result |
| LiveBench ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 41.53% ↗Effort: max · single run · equal mean of 7 suitesIndependent result | 51.47% ↗Effort: max · single run · equal mean of 7 suitesIndependent result |
| LiveBench Design ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 58.4% ↗Effort: max · single runIndependent result | 60.7% ↗Effort: max · single runIndependent result |
| LiveBench Research ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 44.4% ↗Effort: max · single runIndependent result | 70.9% ↗Effort: max · single runIndependent result |
| LiveBench Browser Use ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 4% ↗Effort: max · single runIndependent result | 4% ↗Effort: max · single runIndependent result |
| LiveBench Coding ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 49% ↗Effort: max · single runIndependent result | 65.5% ↗Effort: max · single runIndependent result |
| LiveBench Agentic ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 13.3% ↗Effort: max · single runIndependent result | 15.8% ↗Effort: max · single runIndependent result |
| LiveBench Data Analysis ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 53.3% ↗Effort: max · single runIndependent result | 64.6% ↗Effort: max · single runIndependent result |
| LiveBench Reasoning ↗2.0 · 2026-10-14 (dataset 2026-09-02) | 68.3% ↗Effort: max · single runIndependent result | 78.8% ↗Effort: max · single runIndependent result |
Specifications | ||
| Context window | 262.1K | 1M |
| Max output | 16.4K | — |
| Parameters | 31B | 120B |
| Tool calling | Supported | Not reported |
| Structured output | Supported | Not reported |
| Image input | Supported | Supported |
| Video input | Supported | Not reported |
| Reasoning | Supported | Supported |
| Open weights | Supported | Supported |