GPT-5.5

2026-04-23 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI launches GPT-5.5, its most capable model of the time for code, research, computer use and documents, with the same per-token latency as GPT-5.4.

What changes

  • Marked gains in agentic coding, computer use, knowledge work and scientific research.
  • Uses significantly fewer tokens than GPT-5.4 for the same Codex tasks.
  • GPT-5.5 Pro for hard problems, with more compute.
  • Cyber-permissive models made available through Trusted Access for Cyber, starting with Codex.
  • Default reasoning effort: medium; caching only works there with extended retention.

### Specifications - Context: 1,050,000 tokens (400,000 in Codex), maximum output 128,000 tokens, text and image input. - Knowledge cutoff: December 1, 2025. Effort: none, low, medium (default), high, xhigh. - Availability: ChatGPT (Plus, Pro, Business, Enterprise), Codex, then API (Responses, Chat Completions, Batch). - Identifiers: gpt-5.5, gpt-5.5-2026-04-23, gpt-5.5-pro. ### Pricing | Model | Input | Cached input | Output | | --- | --- | --- | --- | | gpt-5.5 | $5.00 | $0.50 | $30.00 | | gpt-5.5-pro | $30.00 | n/a | $180.00 | Price per million tokens (USD). Beyond 272,000 input tokens: 2x input, 1.5x output. Batch and Flex at half price, Priority at 2.5x. ### Benchmarks | Evaluation | GPT-5.5 | GPT-5.4 | GPT-5.5 Pro | | --- | --- | --- | --- | | Terminal-Bench 2.0 | 82.7% | 75.1% | n/a | | SWE-Bench Pro (public) | 58.6% | 57.7% | n/a | | GDPval (wins or ties) | 84.9% | 83.0% | 82.3% | | OSWorld-Verified | 78.7% | 75.0% | n/a | | BrowseComp | 84.4% | 82.7% | 90.1% | | GPQA Diamond | 93.6% | 92.8% | n/a | | Humanity's Last Exam (with tools) | 52.2% | 52.1% | 57.2% | | FrontierMath level 4 | 35.4% | 27.1% | 39.6% | | ARC-AGI-2 (verified) | 85.0% | 73.3% | n/a | Scores published by the vendor. ### Key takeaways The price per token doubles compared to GPT-5.4 on input ($5 versus $2.50), for lower token consumption. It is the reference for office agents until the arrival of GPT-5.6.

Links

Tags : LLM, OpenAI, GPT-5, Raisonnement, Agent, Codex