GPT-5.2 Medium
| Released | 2025-12-11 |
|---|---|
| Price per 1M | $1.75 in · $14.00 out |
| Overall score | 55.5 · rank 193 |
| Context window | 400,000 tokens |
| Provider | OpenAI |
| Weights | Proprietary |
| Configuration | Medium |
Scores
| Dimension | Score | Rank | Benchmarks |
|---|---|---|---|
| Overall: Balanced | 55.5 | 193 | 13 |
| Overall: Coding | 55.3 | 187 | 8 |
| Overall: Reasoning | 55.2 | 188 | 10 |
| Overall: Agentic work | 55.3 | 195 | 8 |
| Agents | 55.1 | 149 | 2 |
| Reasoning | 55.8 | 192 | 6 |
| Math | 57.1 | 95 | 3 |
| Knowledge | 50.6 | 125 | 1 |
| General | 59.0 | 73 | 1 |
| Value vs price | +0.7 | 221 |
Evidence
Agents 55.1 ±1.9
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Terminal-Bench 2.0 | 64.9% | Epoch AI | 2025-12-24 |
| Remote Labor Index | 2.5% | Epoch AI | 2026-11-11 |
Reasoning 55.8 ±1.8
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Extended NYT Connections | 60.6% | Lech Mazur | 2026-11-10 |
| ARC-AGI-2 | 26.7% | Epoch AI | 2026-11-11 |
| ARC-AGI-2 (ARC Prize verified) | 26.7% | ARC Prize | 2026-11-11 |
| GPQA Diamond | 87.9% | Epoch AI | 2025-12-11 |
| ARC-AGI-1 (ARC Prize verified) | 72.7% | ARC Prize | 2026-11-11 |
| ARC-AGI-1 | 72.7% | Epoch AI | 2026-11-11 |
| Chess Puzzles | 36.0% | Epoch AI | 2026-10-13 |
| Mystery Game Puzzles | 22.0% | Epoch AI | 2026-08-28 |
Math 57.1 ±2.1
| Benchmark | Result | Source | Date |
|---|---|---|---|
| FrontierMath Tiers 1-3 (2025 set) | 36.9% | Epoch AI | 2025-12-11 |
| FrontierMath Tier 4 (2025 set) | 16.7% | Epoch AI | 2025-12-11 |
| OTIS Mock AIME 2024-2025 | 93.9% | Epoch AI | 2025-12-11 |
Knowledge 50.6 ±3.0
| Benchmark | Result | Source | Date |
|---|---|---|---|
| SimpleQA Verified | 32.7% | Epoch AI | 2026-08-27 |
General 59.0 ±5.1
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Creative Story-Writing (Lech Mazur) | 54.0% | Lech Mazur | 2026-10-23 |
Results not used in the score
| Benchmark | Result | Source | Why not used |
|---|---|---|---|
| Terminal-Bench 2.0 | 62.9% | Epoch AI | Lower duplicate run on the same benchmark |
Names matched to this model
gpt-5-2-medium