Full Breakdown
OpenAI Launches GPT-5.5, an Agentic Model for Coding, Knowledge Work, and Research
4/25/2026, 3:29:10 AM
Launch Overview
On April 23 2026 OpenAI released GPT-5.5, a large-language-model designed to execute multi-step tasks across code, spreadsheets, calendars, and scientific tools with minimal direction. The model is immediately available to paid ChatGPT and Codex users (Plus, Pro, Business, Enterprise); API access will follow after additional security reviews.
Competitive Landscape
GPT-5.5 arrives seven weeks after GPT-5.4 and amid a heated race with Anthropic’s Claude Mythos and Google’s Gemini series. Both rivals are preparing IPOs in 2026, competition for enterprise AI services in coding, cybersecurity, and research.
Capabilities
OpenAI’s tests show GPT-5.5 surpassing GPT-5.4 on several benchmarks:
- Terminal-Bench 2.0 (coding): 82.7 % vs. 75.1 % (GPT-5.4)
- GDPval (knowledge work): 84.9 % vs. 78.0 % (Claude Opus 4.7)
- FrontierMath Tier 1-3: 51.7 % vs. 43.8 % (Claude Opus 4.7)
The model uses fewer tokens per task and matches GPT-5.4’s per-token latency while supporting a 1-million-token context window.
Pricing & Availability
ChatGPT and Codex subscribers pay $5 per million input tokens and $30 per million output tokens; the Pro tier costs $30/$180. GPT-5.5 runs on NVIDIA GB200/GB300-NVL72 hardware, with internal load-balancing heuristics improving token-generation speed by over 20 %.
Safety & Criticism
OpenAI classifies GPT-5.5 as “High” risk for cybersecurity and biology/chemistry, applying stricter content filters and a “Trusted Access for Cyber” program for vetted researchers. Critics point to the higher price, friction from safety classifiers, and benchmark gaps—e.g., Gemini 3.1 Pro scores 85.9 % on BrowseComp versus GPT-5.5’s 84.4 %.
Internal Use Cases
OpenAI’s finance team processed 24,771 K-1 tax forms (71,637 pages) two weeks faster than prior year using Codex. An internal variant discovered a new proof about Ramsey numbers, later verified in the Lean theorem prover.
Conflicting Reports
Sources differ on GPT-5.5’s ranking in tool-use benchmarks and on the API launch timeline, with some reports promising “soon” and others citing “outstanding security requirements.”
Verbatim Quotes
- “It just goes and kind of figures it out, deals with ambiguity,” — Greg Brockman, Co-founder and President, OpenAI
- “It can look at an unclear problem and figure out just what needs to happen next.” — Greg Brockman, Co-founder and President, OpenAI
- “I feel more productive because the challenge transitions from figuring out the details of the implementation, the low-level abstractions, to the higher-level goals,” — Jakub Pachocki, Chief Scientist, OpenAI
- “We evaluated this model across our full suite of safety and preparedness frameworks, worked with internal and external redteamers, added targeted testing for advanced cybersecurity and biology capabilities, and collected feedback on real use cases from nearly 200 trusted early-access partners before release,” — OpenAI
Outlook
OpenAI intends to open GPT-5.5 to its Responses and Chat Completions APIs within the next few months, contingent on meeting security prerequisites. The company frames the model as a step toward AI systems that can autonomously accelerate research and move closer to artificial general intelligence.
