OpenAI launches GPT-5.5 for agentic tasks
OpenAI says GPT-5.5 is its smartest and most intuitive model yet, built for real work across coding, research, analysis, documents, spreadsheets, and software use. It is rolling out now in ChatGPT and Codex, with API access promised soon after safety and safeguard work. The pitch is straightforward: better task understanding, fewer tokens per task, and per-token serving speed that matches GPT-5.4, plus API pricing of $5 per 1M input tokens and $30 per 1M output tokens with a 1M context window.
The core story is not just capability, but efficiency: GPT-5.5 is being positioned as a model that understands intent earlier and needs less hand-holding, which is exactly what matters for agentic workflows.
- –Matches GPT-5.4 per-token latency while reportedly using fewer tokens to finish tasks, which is a meaningful cost and UX improvement.
- –The strongest use case is messy, multi-step work: coding, research, analysis, and tool-heavy tasks where models usually need babysitting.
- –The rollout is already live in ChatGPT and Codex, but API users still have to wait for the safety and safeguards process to finish.
- –The pricing is aggressive for a frontier model, especially with a 1M context window, but the real differentiator is likely task efficiency rather than raw benchmark bragging rights.
DISCOVERED
3h ago
2026-04-24
PUBLISHED
17h ago
2026-04-23
RELEVANCE
AUTHOR
sama