GPT-5.5 (Medium): benchmark results
GPT-5.5 evaluated at the medium reasoning-effort setting. Provider: OpenAI. Released 2026-04-23. Access: API.
Aggregate Rating 1807 ± 9, rank #127 of 2168 rated models, from 211 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| DexHoldem Agentic Perception - Community Cards | 61.5 | Field accuracy (%; visible community cards as an order-insen | 100 |
| DexHoldem Agentic Perception - Current Bet Chips | 45.8 | Field accuracy (%; exact per-denomination chip counts of the | 100 |
| DexHoldem Agentic Perception - Field Average | 66.8 | Field accuracy (%; unweighted mean of the eight field accura | 100 |
| DexHoldem Agentic Perception - Loop Stage | 72.2 | Field accuracy (%; loop stage among seven values such as act | 100 |
| DexHoldem Agentic Perception - Robot Chip Inventory | 62.5 | Field accuracy (%; exact per-denomination count of the robot | 100 |
| Facet-Probe - Mixed-Modality Order | 24 | Flip rate (% of items whose normalized answer changes across | 100 |
| Facet-Probe - Option Order | 9 | Flip rate (% of items whose normalized answer changes across | 100 |
| HalluWorld - Chess (No FEN) | 0.3 | Hallucination rate (%) over the seven HalluWorld-Chess probe | 100 |
| HalluWorld - Grid Compound | 2.7 | Hallucination rate (%) on the cross-category compound probes | 100 |
| HalluWorld - Terminal | 5.9 | Hallucination rate (%) over the 529 HalluWorld-Terminal prob | 100 |
| LiveCodeBench-Plus | 62.6 | Pass@1 (%; mean of 4 attempts) | 100 |
| NOLLI - Average | 84.9 | Accuracy (%; 25-task macro average) | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-5-medium · How It Works · Data refreshed daily, snapshot 2026-10-20.