Hybrid reasoning MoE trained on 36T tokens across 119 languages
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 22 models released within six months of it, across 7 benchmarks where scores spread out.
| Benchmark | Score | Rank |
|---|---|---|
MATH Competition-level mathematics problems | 85.7% | #28 / 50 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 88.4% | #33 / 54 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 85% | #44 / 61 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 78.2% | #49 / 62 |
TerminalArtificial Analysis Agentic terminal coding tasks requiring multi-step execution | 13.6% | #51 / 58 |
GPQA PhD-level science questions even experts struggle with | 81.1% | #70 / 96 |
HLEArtificial Analysis Challenging multidisciplinary questions evaluated by Artificial Analysis | 15.9% | #81 / 100 |
| Benchmark / source | Score |
|---|---|
HLE Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 15.9% |
Terminal-Bench 2.1 Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 12% |
Terminal-Bench 4.0 Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 0% |
SciCode Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 41.4% |
AA-LCR Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 72% |
CritPt Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 0% |
IFBench Artificial AnalysisQwen3 235B A22B 2507 (Reasoning) | 51.2% |