- OpenAI GPT-5.5 introduces stronger reasoning and more autonomous, agentic AI capabilities.
- Improved coding, research, and knowledge work performance with higher benchmark scores.
- Faster, more efficient model with better token usage and enhanced safety safeguards.
OpenAI has unveiled GPT-5.5, its latest AI model built to handle complex, multi-step tasks more independently. According to the company, OpenAI GPT-5.5 represents a broader development toward AI systems that can not only respond to prompts but also plan, execute, and refine work across different tools.
The model is rolling out to ChatGPT and Codex users across paid tiers, with API access expected soon.
OpenAI GPT-5.5 brings a shift toward more independent AI workflows
GPT-5.5 is built to better understand user intent, even when instructions are incomplete or unstructured. Instead of requiring step-by-step guidance, it can break tasks into smaller steps, use tools where needed, verify results, and continue working until the task is finished.
This approach makes it more practical for real-world use cases like coding, research, document creation, and data analysis. OpenAI describes this as a move toward agentic AI, where systems can take on extended tasks with less direct supervision.
Stronger coding performance and benchmark gains
OpenAI reports consistent improvements across coding and tool-use benchmarks. On Terminal-Bench 2.0, which measures performance on complex command-line workflows, GPT-5.5 scored 82.7%, up from 75.1% in GPT-5.4.
It also improved on internal evaluations like Expert-SWE, which simulate long, real-world engineering tasks. On SWE-Bench Pro, the model reached 58.6%, showing better performance in resolving real GitHub issues.
Compared with competing systems such as Claude Opus 4.7 and Gemini 3.1 Pro, GPT-5.5 delivers strong results across coding, reasoning, and tool-use benchmarks.
Key benchmark snapshot
| Evaluation | GPT-5.5 | GPT-5.4 | Claude Opus 4.7 | Gemini 3.1 Pro |
|---|---|---|---|---|
| Terminal-Bench 2.0 | 82.7% | 75.1% | 69.4% | 68.5% |
| GDPval (wins/ties) | 84.9% | 83.0% | 80.3% | 67.3% |
| OSWorld-Verified | 78.7% | 75.0% | 78.0% | — |
| Toolathlon | 55.6% | 54.6% | — | 48.8% |
| FrontierMath (Tier 1–3) | 51.7% | 47.6% | 43.8% | 36.9% |
| CyberGym | 81.8% | 79.0% | 73.1% | — |
These results point to steady improvements, particularly in tasks that require reasoning across multiple steps and tools.
More capable in knowledge work and automation
GPT-5.5 also performs strongly in professional and business-related tasks. On GDPval, which evaluates AI across 44 occupations, it achieved 84.9%, outperforming its predecessor.
The model is better at turning unstructured inputs into organized outputs—such as reports, spreadsheets, and presentations. OpenAI says internal teams are already using it to automate workflows, analyze large datasets, and reduce manual effort.
It also shows improved ability to operate software environments, scoring 78.7% on OSWorld-Verified, a benchmark that tests real-world computer interaction.
Early progress in scientific and technical research
In research-focused benchmarks, GPT-5.5 demonstrates stronger multi-step reasoning. It improved on GeneBench and achieved 80.5% on BixBench, which focuses on bioinformatics tasks.
OpenAI says the model is better at navigating the full research cycle—exploring ideas, testing assumptions, and interpreting results. In internal experiments, a version of GPT-5.5 contributed to a mathematical proof related to Ramsey numbers, later verified using formal methods.
Improved efficiency without slowing down
Despite its higher capability, GPT-5.5 maintains similar response speeds to GPT-5.4. OpenAI says the model matches its predecessor’s latency while delivering more accurate and complete outputs.
It is also more token-efficient, meaning it can complete tasks using fewer computational resources, an important factor for both cost and scalability.
Pricing and availability
GPT-5.5 is currently available to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. A more advanced version, GPT-5.5 Pro, is offered for users who need higher accuracy and deeper reasoning.
API pricing (announced)
- GPT-5.5:
- $5 per 1M input tokens
- $30 per 1M output tokens
- GPT-5.5 Pro:
- $30 per 1M input tokens
- $180 per 1M output tokens
The model supports up to a 1 million token context window in API usage, along with options for batch and priority processing.
Keep up with the tech that actually matters.
Add us as a preferred source on Google Search for quicker access to our coverage.
Safety, safeguards, and controlled rollout
OpenAI says GPT-5.5 includes its most advanced safety measures to date. The model underwent extensive testing, including internal evaluations, external red teaming, and targeted assessments in cybersecurity and biology.
The company has implemented stricter controls for sensitive use cases and improved monitoring systems to reduce misuse. At the same time, it is expanding access for verified users working on defensive applications such as cybersecurity.
Rather than relying on a single standout feature, GPT-5.5’s progress comes from a number of improvements, that is, better reasoning, stronger tool use, and the ability to stay on task longer. As adoption grows, its real-world impact will depend on how effectively it integrates into everyday workflows while maintaining responsible use.
