GPT-5.5 (High): benchmark results

GPT-5.5 evaluated at the high reasoning-effort setting. Provider: OpenAI. Released 2026-04-23. Access: API.

Unified ELO 1736 ± 1, rank #45 of 1761 rated models, from 125 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Buyout Game (Lechmazur)1975.4Bradley-Terry Rating100
FutureX49.19Overall Score (latest week, %)100
OpenCompass Code - Comprehensive96.5Score (%)100
OpenCompass LLM - Code96.5Score (%)100
PACT (Lechmazur)1607PACT Bilateral Rating100
Shader Benchmark242.2Avg judge score (0-500, render fails count 0)100
SlopCodeBench28.06Isolated Solved (%)100
SuperCLUE-LongContext - 1M Overall91.98Score100
SuperCLUE-LongContext - 256K Overall98.24Score100
UGI - Natural Intelligence79.27NatInt Score100
Vals AI Terminal-Bench 2.073.2Accuracy (%)100
AA Long Context Reasoning84.33Accuracy (%)99.2

Interactive version: theaggregate.ai/model?slug=gpt-5-5-high · How It Works · Data refreshed daily, snapshot 2026-09-05.