Frontier reasoning model for computer use, coding, science, and professional work. Rolled out in phases starting September 3. Independent evaluations include SciCode and MMMU-Pro.
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 59 models released within six months of it, across 4 benchmarks where scores spread out.
| Benchmark | Score | Rank |
|---|---|---|
GPQAArtificial Analysis PhD-level science questions even experts struggle with | 96.1% | #1 / 95 |
ARC-AGIARC Prize Novel reasoning tasks requiring fluid intelligence | 97.9% | #3 / 39 |
HLEArtificial Analysis Challenging multidisciplinary questions evaluated by Artificial Analysis | 54.7% | #7 / 95 |
| Benchmark / source | Score |
|---|---|
HLE Artificial AnalysisGPT-6 Astra (Max) | 54.7% |
MMMU-Pro Artificial AnalysisGPT-6 Astra (Max) | 86.9% |
Terminal-Bench 2.1 Artificial AnalysisGPT-6 Astra (Max) | 88.4% |
Terminal-Bench 4.0 Artificial AnalysisGPT-6 Astra (Max) | 59.1% |
SciCode Artificial AnalysisGPT-6 Astra (Max) | 56.5% |
AA-LCR Artificial AnalysisGPT-6 Astra (Max) | 80.7% |
CritPt Artificial AnalysisGPT-6 Astra (Max) | 31.7% |
ITBench SRE Artificial AnalysisGPT-6 Astra (Max) | 48.6% |