Originally published at news.skila.ai
OpenAI shipped GPT-5.5 on April 23, 2026. Six weeks after GPT-5.4.
That cadence is the real story. Frontier labs used to ship one major model a year. OpenAI just compressed the release cycle to 42 days, and the pricing tells you why: GPT-5.5 costs 6x more per output token than GPT-5.4, and GPT-5.5 Pro costs more than any model OpenAI has ever sold.
If you built your cost model on GPT-5.4 last month, it just broke. Here is what actually changed and what you should do about it.
What GPT-5.5 Actually Does
The one-line pitch from Sam Altman: GPT-5.5 is the first OpenAI model that feels like a real agent. You give it a task with limited instructions, it figures out the rest.
In practice, that means four things the prior generation could not do reliably:
- Operating software end-to-end. GPT-5.5 can drive a spreadsheet, a document editor, and a browser in a single task. Internal OpenAI evals (codename 'Spud' during development) reportedly showed near-human scores on computer-use benchmarks.
- Deeper research. It pulls sources, reads them, synthesizes across dozens of tabs, and produces a structured report without the usual "here are 40 links" cop-out.
- Better coding. OpenAI claims gains across SWE-bench and terminal-use benchmarks. Early community numbers landed between GPT-5.4 and Claude Opus 4.5's 80.9% on SWE-bench verified — good but not the new king.
- Autonomous task-switching. The model decides when to open a browser vs a code interpreter vs a file editor. No router. No orchestrator. Just the model.
That last point is the subtle one. Every agentic framework shipped in 2025 assumed a human had to wire up the tool routing. GPT-5.5 bundles the routing into the model itself.
The Pricing Is the Headline
Here is what OpenAI confirmed at launch:
- GPT-5.5 standard API: $5 per million input tokens, $30 per million output tokens.
- GPT-5.5 Pro API: $30 per million input tokens, $180 per million output tokens.
Compare that to GPT-5.4, which launched on March 5, 2026 at roughly $2.50 input / $5 output per million tokens. Output pricing jumped 6x in six weeks. Pro pricing is in a league of its own.
Anthropic's Claude Opus 4.5 sits at $15 input / $75 output per million tokens. GPT-5.5 standard undercuts Opus 4.5 on input but charges 40% of Opus 4.5's output rate. GPT-5.5 Pro is $180/M output — 2.4x Opus 4.5.
The 6-Week Cadence Is a Signal
OpenAI used to ship a new frontier model every 6-12 months. GPT-5.4 on March 5, 2026. GPT-5.5 on April 23, 2026. Six weeks.
If you are evaluating AI tooling for a 12-month roadmap, assume at least three more OpenAI launches in 2026.
Who Gets GPT-5.5 Today
ChatGPT Plus, Pro, Business, and Enterprise users get GPT-5.5 immediately. The API rolls out "very soon" — OpenAI wants safeguards figured out for the computer-use capabilities first. Expect a 2-4 week gap.
How It Stacks Against Claude Opus 4.5
| Metric | GPT-5.5 Standard | Claude Opus 4.5 |
|---|---|---|
| SWE-bench verified | ~high-70s | 80.9% |
| Computer use | Reliable, no wrapper needed | Beta |
| Context window | 200K | 200K |
| Output price | $30/M | $75/M |
| API available | In weeks | Now |
Practical answer: if you are picking the hardest coding problems today, Opus 4.5 still wins on benchmarks. If you want an agent that drives software end-to-end with minimal scaffolding, GPT-5.5 is the first model that does that reliably.
Three Immediate Implications for Developers
- Cost modeling has to be redone. If you priced a product on GPT-5.4 output at $5/M, you cannot port that pricing to GPT-5.5.
- Agentic capability is the new moat. Raw benchmark scores matter less than the ability to operate a browser, a spreadsheet, and a terminal without a wrapper.
- Switching costs are dropping. GPT-5.5, Opus 4.5, and Gemini 2.5 Pro all expose the same OpenAI-compatible API surface. You can swap model providers in one config change.
Verdict
GPT-5.5 is genuinely better at agentic work than GPT-5.4. The 6-week cadence shows OpenAI is in war-footing mode. The pricing is the highest OpenAI has ever charged and will force a round of cost-model rewrites across every production AI app.
If you are a ChatGPT Business customer, test the computer-use features on your hardest workflow now. If you are an API developer, wait 2-4 weeks for the rollout and re-estimate your token spend before migrating.
For the full breakdown with FAQ, read the original at news.skila.ai.
Top comments (0)