GPT-5.2: benchmark results
OpenAI's standard GPT-5.2 model for general-purpose reasoning, coding, and writing. Provider: OpenAI. Released 2025-12-11. Access: API.
Aggregate Rating 1755 ± 2, rank #110 of 1656 rated models, from 3659 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AEC-Bench (Nomic Tools) - Detail Technical Review | 85.7 | Mean reward (0-100) from the task verifier on the 14 detail- | 100 |
| AEC-Bench - Detail Technical Review | 60.7 | Mean reward (0-100) from the task verifier on the 14 detail- | 100 |
| AEC-Bench - Drawing Navigation | 83.3 | Mean reward (0-100) from the task verifier on the 12 drawing | 100 |
| ALL Bench Multimodal | 86.7 | Average Numeric VLM Score (%) | 100 |
| AMA-Bench | 69.83 | Average Score (%) | 100 |
| AMA-Bench - Game | 81.13 | Average Score (%) | 100 |
| AgencyBench V2 - Backend | 61 | Average Task Score (%) | 100 |
| AgencyBench V2 - Code | 50.7 | Average Task Score (%) | 100 |
| AgencyBench V2 - Overall | 56.5 | Average Task Score (%) | 100 |
| Agent2 RL-Bench - GSM8K | 89.46 | GSM8K test accuracy (%) of the Qwen3-8B-Base model after the | 100 |
| AgentNoiseBench (VitaBench, User Noise) | 45 | Success on a stratified 25 percent sample of VitaBench tasks | 100 |
| AgentNoiseBench (tau2-bench, Tool Noise) | 61 | Success on a stratified 25 percent sample of tau2-bench task | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-2 · How It Works · Data refreshed daily, snapshot 2026-10-20.