Final specialized Codex release before coding was folded into mainline GPT-5.4. Peak on LiveCodeBench and Terminal-Bench.
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 52 models released within six months of it, across 8 benchmarks where scores spread out.
| Benchmark | Score | Rank |
|---|---|---|
MMLU Tests knowledge across 57 subjects from STEM to humanities | 93% | #4 / 54 |
OSWorld Computer use in real desktop environments | 74% | #4 / 13 |
SWE-bench Real-world GitHub issue resolution | 80% | #6 / 39 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 77.3% | #8 / 58 |
MATH Competition-level mathematics problems | 96% | #15 / 50 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 87.3% | #15 / 62 |
GPQA PhD-level science questions even experts struggle with | 92.6% | #21 / 96 |
HumanEval Coding ability - generating correct Python functions | 93% | #24 / 50 |
ARC-AGI Novel reasoning tasks requiring fluid intelligence | 54% | #26 / 40 |
HLEArtificial Analysis Challenging multidisciplinary questions evaluated by Artificial Analysis | 42.5% | #32 / 100 |
| Benchmark / source | Score |
|---|---|
HLE Artificial AnalysisGPT-5.3 Codex (Xhigh) | 42.5% |
MMMU-Pro Artificial AnalysisGPT-5.3 Codex (Xhigh) | 78.5% |
AA-LCR Artificial AnalysisGPT-5.3 Codex (Xhigh) | 83.3% |
CritPt Artificial AnalysisGPT-5.3 Codex (Xhigh) | 16.9% |
IFBench Artificial AnalysisGPT-5.3 Codex (Xhigh) | 75.4% |