OpenAI's fully retrained flagship. Doubled in API price over GPT-5.4, with a much bigger 1M token context.
GPT-5.5 is not a fine-tune of GPT-5.4 — OpenAI describes it as a from-scratch retraining released on April 23, 2026. The headline numbers tell most of the story: API pricing doubled (input went from $2.50 to $5, output from $15 to $30) and the context window jumped from 272K to a flat 1M tokens. Both moves push GPT-5.5 directly into the same lane as Gemini 3.1 Pro and the higher Anthropic tiers.
The price increase is real but defensible. For most evaluators, GPT-5.5 lands meaningfully above GPT-5.4 on long-context recall, multi-step tool use, and code generation against larger repos. If your workload is a quick classification call, GPT-5.4 mini at $0.25/$2 is still the right pick — GPT-5.5 is built for the harder tier where smarter beats cheaper.
GPT-5.5 is a generalist. It handles long documents, agentic tool calling, vision, structured output, and code with roughly equal competence — which is rare. The 1M context is the most quietly important upgrade: you can drop an entire codebase, a book, or a multi-hour transcript in without splitting it across calls and losing coherence. The model is also faster than GPT-5.5 pro mode at most query lengths, which matters once you're shipping it inside a product.
Where it still trails: pure math and theorem-style reasoning, where DeepSeek V4-Pro and Claude Opus 4.7 are stronger; and image generation, which OpenAI keeps in a separate model.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| GPT-5.5 | $5 | $30 | 1M |
| GPT-5.4 | $2.50 | $15 | 272K |
| Claude Opus 4.7 | $5 | $25 | 200K |
| Gemini 3.1 Pro | $2 | $12 | 1M |
Opus 4.7 matches GPT-5.5 on input and undercuts it on output, but the context window is 5× smaller. Gemini 3.1 Pro is materially cheaper but pricing rises above 200K tokens. Choose GPT-5.5 when you want the most general-purpose answer, not the cheapest one.