GPT-5.5¶
OpenAI announced GPT-5.5 on April 23, 2026, positioning it as a fundamental redesign of how AI interacts with operating systems and professional software. GPT-5.5 retakes the benchmark lead from Anthropic's Claude Opus 4.7 (released the previous week) across the majority of publicly available evaluations, with a particular focus on agentic computer use, coding, and scientific research (VentureBeat, 2026-04-23).
Availability¶
At launch, GPT-5.5 is available to ChatGPT subscribers at the Plus ($20/month), Pro ($100–200/month), Business, and Enterprise tiers. API access was not available at launch and was described as "coming very soon," with OpenAI citing different safety requirements for API vs. consumer deployment. GPT-5.5 is available in two variants:
- GPT-5.5 — general-purpose flagship for intelligence tasks
- GPT-5.5 Pro — enhanced precision and specialised logic for high-stakes environments (legal research, data science, advanced analytics); starts at the Pro subscription tier
What's New¶
OpenAI describes GPT-5.5 as focused on agentic performance: handling messy, multi-part tasks autonomously without step-by-step prompting. Unlike predecessors that required granular guidance to avoid hallucinating a path forward, GPT-5.5 is designed to look at an unclear problem and determine what needs to happen next.
The key technical improvement is hardware-software co-design efficiency. GPT-5.5 was served on NVIDIA GB200 and GB300 NVL72 systems with custom heuristic algorithms — written by the model itself — that partition and balance work across GPU cores, reportedly increasing token generation speeds by over 20% while matching GPT-5.4's per-token latency. The model is simultaneously more capable and no slower than its predecessor.
GPT-5.5 Thinking mode in ChatGPT provides smarter, more concise answers by giving the model internal compute time to verify assumptions before responding. This is particularly visible on Expert-SWE (internal OpenAI benchmark for long-horizon coding tasks with a 20-hour human median), where GPT-5.5 outperformed GPT-5.4 using significantly fewer tokens.
Benchmark Comparison¶
| Benchmark | GPT-5.5 | Claude Opus 4.7 | Gemini 3.1 Pro | Mythos Preview* |
|---|---|---|---|---|
| Terminal-Bench 2.0 | 82.7 | 69.4 | 68.5 | 82.0 |
| ARC-AGI-2 (Verified) | 85.0 | 75.8 | 77.1 | — |
| OSWorld-Verified | 78.7 | 78.0 | — | 79.6 |
| BrowseComp | 84.4 | 79.3 | 85.9 | 86.9 |
| SWE-bench Pro (Public) | 58.6 | 64.3 | 54.2 | 77.8 |
| Humanity's Last Exam (no tools) | 43.1 | 46.9 | — | 56.8 |
| FrontierMath Tier 1–3 | 51.7 | 43.8 | 36.9 | — |
*Mythos Preview is not generally available; restricted to trusted partners and government agencies.
GPT-5.5 leads on computer use, agentic tasks, and complex mathematics. Claude Opus 4.7 leads on pure software engineering (SWE-bench) and multidisciplinary reasoning without tools. Google Gemini 3.1 Pro leads in academic reasoning and financial analysis in three specific benchmarks.
Pricing¶
OpenAI doubled API pricing compared to GPT-5.4, with GPT-5.5 Pro at 6x GPT-5.4's price:
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GPT-5.4 | $2.50 | $15.00 |
| GPT-5.5 | $5.00 | $30.00 |
| GPT-5.5 Pro | $30.00 | $180.00 |
OpenAI notes that GPT-5.5 is more token-efficient — it completes the same tasks with fewer tokens than GPT-5.4. A Fast mode in Codex generates tokens 1.5x faster at a 2.5x price premium.
Safety and Licensing¶
Under OpenAI's Preparedness Framework, GPT-5.5 is classified as "High" risk for biological and cybersecurity capabilities. OpenAI has implemented stricter "cyber-risk classifiers" for general users and a "cyber-permissive" licence for verified security professionals responsible for critical infrastructure, allowing fewer refusals for security-related prompts.
API deployments require different safeguards from consumer ChatGPT, which is why API access was delayed at launch. OpenAI is working with government partners on safety and security requirements.
Competitive Context¶
As of the April 23 launch, GPT-5.5 leads on 14 of the reported benchmarks versus 4 for Claude Opus 4.7 and 2 for Gemini 3.1 Pro. However, Anthropic's restricted Claude Mythos Preview model leads on several benchmarks where it is included — but Mythos is not commercially available, so the practical competitive comparison for enterprises remains GPT-5.5 vs. Opus 4.7 vs. Gemini 3.1 Pro.
OpenAI chief scientist Jakub Pachocki stated that the company "still has headroom to train significantly smarter models than this," suggesting GPT-5.5 is not a scaling ceiling.
Limitations¶
This page rests on a single VentureBeat article; the OpenAI announcement blog post was not directly retrieved. Benchmark figures are from OpenAI-provided material rather than independent third-party evaluation, and internal benchmarks (Expert-SWE, GDPval, Investment Banking Modeling Tasks) cannot be independently verified. API pricing and availability details should be confirmed against the official OpenAI pricing page, as pricing structures frequently change post-launch.
Changelog¶
2026-05-19 — Page created from VentureBeat (Type C, 2026-04-23). Confidence 60 (single-source cap). Topics: GPT-5.5 launch, benchmarks, variants, pricing, safety classification.