Model Details
GPT-5.5
Quick answer: GPT-5.5 is OpenAI's June 2026 frontier multimodal model, scoring 95.0% on ARC-AGI, 93.6% on GPQA Diamond, 84.4% on BrowseComp, 81.8% on CyberGym, 47.9% on Agents Last Exam, and 41.4% on Humanity's Last Exam. The current top-tier GPT-5 series model.
At a Glance
Where GPT-5.5 leads
- 95.0% ARC-AGI — near-ceiling abstract reasoning (+1.3pp vs GPT-5.4)
- 93.6% GPQA Diamond — state-of-the-art graduate science
- 81.8% CyberGym — leading cybersecurity benchmark
- 84.4% BrowseComp — strong web research
- 47.9% Agents Last Exam — strong agentic task performance
- Full multimodal capability
Where it lags
- 41.4% HLE — below Gemini 3.1 Pro (46.4%) and GLM-5.2 (54.7%)
- ARC-AGI still below Gemini 3.1 Pro (98.0% vs 95.0%)
- Proprietary
Best for: Maximum OpenAI capability; agentic tasks requiring the strongest GPT-5 model; multimodal frontier use cases.
What GPT-5.5 Is
GPT-5.5 (released June 2026) is OpenAI's most capable model at this release date — the June 2026 increment of the GPT-5 series. It improves on GPT-5.4 across all available benchmarks and adds stronger agentic performance (47.9% Agents Last Exam vs GPT-5.4's 37.3%).
The 81.8% CyberGym score is notable — significantly above GPT-5.4's 79.0% and placing it among the strongest models on security evaluation. The BrowseComp improvement (84.4% vs 82.7%) also indicates stronger web research capability.
Specifications
| Field | Value |
|---|---|
| Organization | OpenAI |
| License | Proprietary (API only) |
| Release date | June 2026 |
| Knowledge cutoff | September 2025 |
| Modality | Multimodal |
Pricing
Available via OpenAI API. Refer to the OpenAI pricing page for current GPT-5.5 rates.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| ARC-AGI | 95.0% | Benchgen evaluation | 2026-06 |
| GPQA Diamond | 93.6% | Benchgen evaluation | 2026-06 |
| BrowseComp | 84.4% | Benchgen evaluation | 2026-06 |
| CyberGym | 81.8% | Benchgen evaluation | 2026-06 |
| Agents Last Exam | 47.9% | Benchgen evaluation | 2026-06 |
| Humanity's Last Exam | 41.4% | Benchgen evaluation | 2026-06 |
GPT-5.5 vs Alternatives
| Model | ARC-AGI | GPQA Diamond | HLE | CyberGym |
|---|---|---|---|---|
| GPT-5.5 | 95.0% | 93.6% | 41.4% | 81.8% |
| GPT-5.4 | 93.7% | 92.8% | 39.8% | 79.0% |
| Gemini 3.1 Pro | 98.0% | 94.3% | 46.4% | 38.8% |
| GLM-5.2 | — | 91.2% | 54.7% | — |
GPT-5.5 vs Gemini 3.1 Pro: GPT-5.5 leads on CyberGym (81.8% vs 38.8%); Gemini leads on ARC-AGI (98% vs 95%) and HLE (46.4% vs 41.4%). Choice depends on task: GPT-5.5 for security/coding, Gemini 3.1 for math/science.
Related Models & Benchmarks
- Compare: GPT-5.4 · Gemini 3.1 Pro · GLM-5.2
- Benchmarks: ARC-AGI · GPQA Diamond · CyberGym · Humanity's Last Exam
Frequently Asked Questions
What is GPT-5.5?
OpenAI's June 2026 frontier multimodal model scoring 95.0% ARC-AGI, 93.6% GPQA Diamond, 81.8% CyberGym, and 41.4% HLE. The current top-tier GPT-5 series model.How does GPT-5.5 compare to Gemini 3.1 Pro?
Both are frontier multimodal models. Gemini 3.1 Pro leads on ARC-AGI (98% vs 95%), AIME 2026 (98.3%), and HLE (46.4% vs 41.4%). GPT-5.5 leads on CyberGym (81.8% vs 38.8%). Choice depends on task requirements.Specs from OpenAI's GPT-5.5 release (June 2026) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.