OpenAI unveils GPT-5.5, designed for agentic work
- Maxime Hiez
- OpenAI
- 28 Apr, 2026
Introduction
OpenAI has just announced GPT-5.5, barely seven weeks after GPT-5.4. The message is clear, GPT-5.5 is not an incremental update but “a new class of intelligence for real work”. The goal is no longer just to generate relevant text, but to understand long and complex tasks, plan steps, use tools, and achieve the final result.
What really changes with GPT-5.5
Significantly enhanced autonomy
GPT-5.5 is designed to work with fewer explicit instructions. Where previous versions required precise framing, this one is capable of :
- Inferring intermediate steps
- Handling ambiguity
- Adjusting its approach during the task
A natively agentic approach
GPT-5.5 is the first model fully retrained since GPT-4.5 and the first to be natively designed for agentic programming.
- Multi-step planning
- Use of tools (code, navigation, files, etc.)
- Self-checking of results
- Iterations until completion
Performance vs. efficiency
An interesting point is that the gains aren’t solely in “raw intelligence”. GPT-5.5 also aims to :
- Reduce the number of tokens required
- Improve the quality of the first result
- Limit the number of iterations
The 4 areas where GPT-5.5 truly makes a difference
Software development (agentic coding)
- Understanding a complex GitHub repository
- Solving complex multi-file tasks
- Generating + fixing + validating
Use of tools (computer use)
Operate a computer independently :
- Browsing the web
- Manipulating files
- Using applications
Knowledge Work
- Analyze data
- Create deliverables (documents, tables, summaries)
Research and Exploration
- Generate hypotheses
- Structure research paths
- Explore complex topics
Benchmarks : GPT-5.5 vs the competition
Here are the results published by OpenAI, compared to competing models :
| GPT-5.5 | GPT‑5.4 | GPT-5.5 Pro | GPT-5.4 Pro | Claude Opus 4.7 | Gemini 3.1 Pro | |
|---|---|---|---|---|---|---|
| Terminal-Bench 2.0 | 82.7% | 75.1% | - | - | 69.4% | 68.5% |
| Expert-SWE (internal) | 73.1% | 68.5% | - | - | - | - |
| GDPval (wins or ties) | 84.9% | 83.0% | 82.3% | 82.0% | 80.3% | 67.3% |
| OSWorld-Verified | 78.7% | 75.0% | - | - | 78.0% | - |
| Toolathon | 55.6% | 54.6% | - | - | - | 48.8% |
| BrowseComp | 84.4% | 82.7% | 90.1% | 89.3% | 79.3% | 85.9% |
| FrontierMath Tier 1–3 | 51.7% | 47.6% | 52.4% | 50.0% | 43.8% | 36.9% |
| FrontierMath Tier 4 | 35.4% | 27.1% | 39.6% | 38.0% | 22.9% | 16.7% |
| CyberGym | 81.8% | 79.0% | - | - | 73.1% | - |
GPT-5.5 Thinking vs GPT-5.5 Pro
GPT-5.5 comes in two versions, depending on the level of task complexity :
- GPT-5.5 Thinking
Unlocks faster help for tough problems, with smarter, more concise answers to help you get ahead with complex tasks. Available to Plus, Pro, Business, and Enterprise subscribers in ChatGPT and Codex.
- GPT-5.5 Pro
Presented as a research partner for challenging questions where accuracy matters more than speed. Testers describe it as “a real leap forward in the difficulty and quality of work that ChatGPT can handle, with latency improvements that make it far more practical for demanding tasks”. Available for Pro, Business, and Enterprise only.
API pricing
The API price of GPT-5.5 is twice as high as GPT-5.4, but OpenAI justifies this extra cost with better token efficiency (fewer tokens to accomplish the same task) and significantly superior intelligence.
| Version | Input | Cached Input | Output |
|---|---|---|---|
| GPT‑5.5 | 5$ / 1M tokens | 0.50$ / 1M tokens | 30$ / 1M tokens |
| GPT‑5.4 | 2.50$ / 1M tokens | 0.25$ / 1M tokens | 15$ / 1M tokens |
note
What this means for businesses
OpenAI’s message is strategic, in a context where agentic AI is becoming the new competitive landscape, GPT-5.5 positions OpenAI as the benchmark for enterprise computer “work workflows”. For organizations using Codex, ChatGPT Enterprise, or the OpenAI API in their workflows, GPT-5.5 opens up concrete possibilities :
- Less human supervision of repetitive or multi-step tasks.
- Improved code quality and architectural understanding in complex projects.
- Increased productivity on knowledge work tasks (analysis, writing, spreadsheets).
- Enhanced research for R&D teams, with the ability to generate and test hypotheses more autonomously.
Conclusion
GPT-5.5 marks a turning point in OpenAI’s trajectory, moving from an assistant that responds to a system that acts. With its native agentic capabilities, its million-token context window, its unparalleled performance on the computer use and knowledge work benchmarks, and its maintained latency despite the leap in capacity, GPT-5.5 represents a concrete step towards AI capable of handling true end-to-end workflows. The release cadence is accelerating, the competition with Anthropic is intensifying, and the real winner is the productivity of the teams that know how to leverage these tools.
Sources
Did you enjoy this post ? If you have any questions, comments or suggestions, please feel free to send me a message from the contact form.
Don’t forget to follow us and share this post.
