Models
GPT (GPT-5.5)
OpenAI's flagship model family — the engine behind ChatGPT's default experience.
Model Facts
Provider
OpenAI
Released
April 23, 2026
Context window
1M tokens (API)
Weights
Closed
MMLU
92.4%
SWE-bench Verified
88.7%
What it is
GPT-5.5 is OpenAI's current flagship model, powering ChatGPT's default experience and the OpenAI API. It's part of a rapid succession of GPT-5.x releases — OpenAI has been shipping updates to this line every few months rather than waiting years between major versions, following the original GPT-4o launch that made the family natively multimodal (text, image, and audio in one model).
Where these numbers come from
The benchmarks above are OpenAI's officially published figures for GPT-5.5, confirmed April 23, 2026: 92.4% on MMLU and 88.7% on SWE-bench Verified, with a 1M-token context window available via the API (400K in Codex). OpenAI has since previewed a newer GPT-5.6 generation (three variants: Sol, Terra, and Luna) starting June 26, 2026, but as of this writing it remains a gated preview limited to a small number of approved companies, with no official coding-benchmark numbers published yet. Update this entry via the admin dashboard once GPT-5.6 (or its successor) reaches general availability with confirmed benchmarks.
Notable traits
- Ships in multiple tiers by cost/speed trade-off (a pattern OpenAI has continued into the 5.6 preview with its Sol/Terra/Luna split) rather than a single one-size-fits-all model.
- Reported to have meaningfully reduced hallucination rates compared to earlier GPT-5 generations.
- Closed weights — available only through OpenAI's API and products, not for local download (see Open Weight vs Closed Weight Models).