Yesterday (April 23), OpenAI released GPT-5.5.
Codename: Spud. It arrives exactly six weeks after GPT-5.4.
If you recall a year ago, OpenAI was releasing a major version every quarter, and everyone thought that was fast enough. Now, a new version every six weeks — the pace has become more like a software iteration cycle.
Pricing first — the most direct signal
| Version | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Standard | $5 | $30 |
| Pro | $30 | $180 |
Pro output is priced at $180/M. Claude Opus 4.7 is $75/M — that's 2.4 times more expensive. This price isn't arbitrary; it indicates that OpenAI's target customers are not individual developers.
As for capabilities, Greg Brockman put it directly:
"What's really special about this model is that it can do more with less guidance. Given a vague problem, it can figure out what to do next on its own."
What can it actually do?
- Directly operate email, calendar, spreadsheets, and software interfaces
- Multi-step tasks across sessions without requiring human oversight
- Conduct online research and output reports with citations
- Improved code debugging and execution capabilities
- Notable improvements in math and scientific research tasks
Brockman also described this as a step toward more agentic and intuitive computing — faster, more precise, and getting things done with fewer tokens.
It has been rolled out to Plus, Pro, Business, and Enterprise users, and is available directly in ChatGPT and Codex.
OpenAI is no longer selling a chat model
This is the real story behind GPT-5.5.
Previously, competitors were competing on reasoning ability, multimodal quality, and context length. Now, OpenAI is selling an agent runtime that can autonomously complete workflows.
Give it a vague instruction, and it completes the entire task before returning. No need for you to design prompts, break down tasks, or supervise every step.
This explains a lot:
- Why Pro pricing is so expensive — enterprises aren't paying for chat; they're paying for labor replacement
- Why a new version every six weeks — it's not about chasing benchmarks; it's about locking in user muscle memory
- Why the emphasis on less guidance — lowering the barrier to entry so non-technical decision-makers can also buy in
What this means for competitors
Frankly, both Anthropic and Google are working in this direction and have products.
Claude Cowork is an enterprise AI agent, and Google Gemini Enterprise Agent Platform was just announced at Cloud Next 2026. Everyone is talking about agentic computing.
GPT-5.5's position is: leverage OpenAI's largest existing user base (ChatGPT) and push agent capabilities directly into it. Not starting from scratch, but adding features on top of existing traffic.
The advantage of this move is clear — Cursor has millions of developer users; OpenAI has hundreds of millions of regular users. The coverage isn't even in the same order of magnitude.
But the pricing is genuinely expensive. $180/M for Pro output has limited appeal for cost-sensitive engineering teams. Coincidentally, DeepSeek V4-Pro output is $3.48/M, and it was also released today (April 24).
How the price war will unfold next is worth watching.
A model every six weeks — what is OpenAI playing at?
To be honest, the six-week release cycle itself is a competitive strategy.
Previously, major model releases were big events requiring buildup, hype, and launch events. Now, OpenAI is treating it more like software iteration: small steps, fast releases, continuously occupying user attention, and leaving competitors with no time to concentrate their response.
You spend three months building a milestone model; they release one every six weeks that's slightly better than the last. Which one makes it harder for users to switch?
This logic isn't new, but in the AI industry, OpenAI is the first to truly execute this rhythm.
As for when GPT-5.6 will arrive — six weeks from now would be around early June.
Sources: OpenAI releases GPT-5.5, bringing company one step closer to an AI 'super app' (TechCrunch, April 23, 2026); CocoLoop, GPT-5.5 Just Shipped April 2026 and OpenAI Finally Built a Real Agent (RoboRhythms, April 23, 2026); OpenAI unveils GPT-5.5, marking new era of autonomous 'digital workers' (Geo.tv, April 23, 2026)