Apache 2.0 dense 31B from the Gemma 4 family. Native vision + audio, 140+ languages. Best open-weight from Google Q2 2026.
SI v3 uses weighted percentages from one Artificial Analysis configuration. Ranking requires all five benchmarks. Partial scores average only available weights. Scores are not comparable to SI v2 or a probability of reaching singularity.
Compared with models released within six months of it.
| Benchmark | Score | Rank |
|---|---|---|
TerminalArtificial Analysis Agentic terminal coding tasks requiring multi-step execution | 36.4% | #37 / 58 |
Arena EloArtificial Analysis Human preference ranking via blind comparisons | 1114 | #51 / 51 |
GPQAArtificial Analysis PhD-level science questions even experts struggle with | 85.7% | #61 / 96 |
HLEArtificial Analysis Challenging multidisciplinary questions evaluated by Artificial Analysis | 23.6% | #75 / 100 |
| Benchmark / source | Score |
|---|---|
HLE Artificial AnalysisGemma 4 31B (Reasoning) | 23.6% |
MMMU-Pro Artificial AnalysisGemma 4 31B (Reasoning) | 73.4% |
Terminal-Bench 2.1 Artificial AnalysisGemma 4 31B (Reasoning) | 43.4% |
Terminal-Bench 4.0 Artificial AnalysisGemma 4 31B (Reasoning) | 0% |
SciCode Artificial AnalysisGemma 4 31B (Reasoning) | 45.5% |
AA-LCR Artificial AnalysisGemma 4 31B (Reasoning) | 69.7% |
CritPt Artificial AnalysisGemma 4 31B (Reasoning) | 1.4% |
ITBench SRE Artificial AnalysisGemma 4 31B (Reasoning) | 37.3% |
IFBench Artificial AnalysisGemma 4 31B (Reasoning) | 75.6% |