GPT-5.5 – Capable and Highly Accurate Model for Real Work
  • Nano Banana
  • Nano Banana 2
  • Seedream 4.0
  • 30+ AI Models
  • GPT-5.5: Smarter decisions, Less guidance required

    GPT- 5.5 is OpenAI's most capable model designed for complex, real-world work. It figures out what needs to happen next, uses tools to get there, and checks its own work along the way.

    Trusted by users from 10,000+ companies

    Additional Capabilities GPT-5.5 Brings

    Apart from headline changes, GPT-5.5 offers smaller improvements that compound into bigger advantages.

    Token efficiency

    Token efficiency

    GPT-5.5 completes the same tasks using significantly fewer tokens than GPT-5.4, delivering more capable results at lower cost in Codex.

    Same-speed serving

    Same-speed serving

    Despite the jump in capability, GPT-5.5 matches GPT-5.4's per-token latency in real-world serving. More intelligence, no wait penalty.

    Computer use

    Computer use

    GPT-5.5 can operate software directly, navigating interfaces, filling forms, and executing multi-step tasks the way a person would.

    Frequently Asked Questions

    Quick answers covering access, pricing, capabilities, and how GPT-5.5 compares to previous models.

    Scientific reasoning

    Scientific reasoning

    Meaningful gains on scientific and technical research workflows, including early-stage drug discovery and experimental debugging tasks.

    Destructive action prevention

    Destructive action prevention

    GPT-5.5 distinguishes its own work from user work and reverts only its own changes, protecting user output even in complex, multi-agent environments.

    Context retention

    Context retention

    Holds context reliably across large systems and long sessions, carrying changes through surrounding codebases without losing track of earlier decisions.

    Document creation

    Document creation

    Improved latency on document and spreadsheet creation tasks, producing complete, structured outputs faster without sacrificing accuracy.

    Cybersecurity access

    Cybersecurity access

    Verified defenders can apply for Trusted Access for Cyber to unlock GPT-5.5's advanced security capabilities with fewer restrictions for legitimate defensive work.

    Codex integration

    Codex integration

    GPT-5.5 is the default model in Codex for all paid tiers, with OpenAI's tuning specifically optimised for agentic coding performance at lower token cost.

    Frontier Performance in Real-world Tasks

    GPT-5.5 leads across the evaluations that mirror how professionals and researchers actually work.

    Excels at Clinician-use Evaluation

    Excels at Clinician-use Evaluation

    GPT-5.5 scores higher than any previous GPT model on HealthBench Professional,designed for clinical use cases, with both higher raw and length-adjusted scores.

    Top Agentic Coding Model

    Top Agentic Coding Model

    GPT-5.5 leads SWE-Bench Verified, placing it at the top of the field for resolving real production software issues autonomously across full codebases.

    Most Accurate GPT Model

    Most Accurate GPT Model

    On conversations where users had previously flagged factual errors, GPT-5.5's individual claims are 23% more likely to be correct, improving precision and trustworthiness.

    Work that Figures Itself Out

    GPT-5.5 understands what a task actually requires before starting, identifies the gaps in ambiguous briefs, and builds a plan without waiting to be walked through it. Early access teams reported saving up to 10 hours of work per week.

    Work that Figures Itself Out

    Research and Analysis at a Different Speed

    GPT-5.5 handles multi-step research workflows autonomously, searching, synthesising, and checking its conclusions against source material. What previously required breaking a brief into a sequence of separate prompts now runs as a single task.

    Research and Analysis at a Different Speed

    Accuracy on Work that Cannot Afford Errors

    GPT-5.5 is more factually reliable than its predecessors, particularly on precision-dependent tasks in finance, law, medicine, and research. Early users at regulated financial institutions reported a step-change in output accuracy.

    Accuracy on Work that Cannot Afford Errors