OpenAI just shipped GPT-5.5. The leaks were real. The Polymarket bets paid off. And the model is live starting today for Plus, Pro, Business, and Enterprise subscribers across ChatGPT and Codex. This is not another incremental point release. GPT-5.5 is the first fully retrained base model OpenAI has shipped since GPT-4.5, and it lands just six weeks after GPT-5.4. That turnaround is the fastest OpenAI has ever moved between flagship releases. Greg Brockman called it "a new class of intelligence for real work." Sam Altman's long-teased super app vision just got the model that actually makes it possible. And the framing from OpenAI today is different from every launch before it. They are not pitching it as a better chatbot. They are pitching it as an agent that finishes tasks without you babysitting every step. That is a different product category. And the entire AI industry is about to feel it.

OpenAI is being explicit about the shift. This is not a chat model with upgrades. This is an agent runtime. Brockman's framing on the press call was direct. You give GPT-5.5 a messy, multi-part task and trust it to plan, use tools, check its work, navigate through ambiguity, and keep going until the task is done. No more stalling at handoff points. No more re-prompting halfway through a workflow. The model is designed to complete end-to-end work.

Here is what the numbers actually show.

82.7% on Terminal-Bench 2.0. This benchmark tests how well a model can operate inside a terminal environment, running commands, debugging scripts, orchestrating pipelines, and fixing its own mistakes. For ML engineers and data scientists who live in terminals, this is the benchmark that matters most. 82.7% puts GPT-5.5 at the top of the pack.

84.9% on GDPval. This measures economic productivity across real world tasks. High scores here mean the model can actually do work that generates value, not just answer questions.

58.6% on SWE-Bench Pro. This evaluates real GitHub issues across four programming languages. Worth noting Claude Opus 4.7 scores 64.3% on the same benchmark, but OpenAI flagged that Anthropic reported signs of memorization on some problems which may affect the comparison. Either way, GPT-5.5 has closed the gap significantly from where GPT-5.4 sat against Claude.

State of the art on the Artificial Analysis Coding Index. And at roughly half the cost of competing frontier coding models according to OpenAI. That pricing matters. If GPT-5.5 delivers comparable coding performance to Claude Opus 4.7 at half the price, enterprise contracts are going to feel that math immediately.

Efficiency improvements are real. OpenAI says GPT-5.5 uses fewer tokens to solve the same problems. Fewer retries. Less glue code between steps. That compounds into actual cost savings for teams running production workflows.

Pricing for the API: Standard GPT-5.5 is $5 per million input tokens and $30 per million output. GPT-5.5 Pro jumps to $30 input and $180 output for serious agent work with long context and heavy token generation. That Pro pricing is genuinely expensive. More than Anthropic's Opus pricing. OpenAI is betting the reliability makes the price worth it.

The scale behind the launch: 900 million weekly active ChatGPT users. 50 million paying subscribers. 9 million paying business users. 4 million active Codex users. OpenAI has the biggest distribution in AI and they just dropped their most capable model into every tier.

Because OpenAI needed this launch more than any launch they have ever done.

Since December 2025 OpenAI has been in internal Code Red mode. Claude closed the capability gap faster than leadership expected. Claude Sonnet 4.6 started winning real world work benchmarks. Claude Opus 4.7 took the top spot on coding evals. Claude Mythos changed the cybersecurity conversation. Claude Design took a run at Figma and moved the stock 7 percent. The narrative across developer communities and enterprise contracts shifted. OpenAI was the consumer brand and Anthropic was the technical leader. Every week that narrative held was another week of lost ground.

GPT-5.5 is OpenAI's answer.

And the framing is deliberate. OpenAI is not trying to win a chat benchmark anymore. They are trying to own the agent category before Anthropic does. That is why Brockman called this a new class of intelligence. That is why the pitch leads with "it finishes tasks." That is why the pricing undercuts Claude on standard tier. OpenAI wants to define what an agent model is before Anthropic gets to.

The cybersecurity framing is the other big signal. When asked if GPT-5.5 has capabilities similar to Claude Mythos, OpenAI's chief research officer Mark Chen said the model shows meaningful gains on scientific and technical research workflows and could help expert scientists make progress including on drug discovery. GPT-5.5 meets "High" cyber risk classification but not "Critical." Translation, it is powerful enough to be dangerous but OpenAI is releasing it widely anyway. That is a different philosophical choice from Anthropic who gated Mythos to 40 defenders. OpenAI is betting on distribution. Anthropic is betting on control.

If you use ChatGPT, GPT-5.5 is already rolling out to your Plus subscription. You will notice it feels different. Less back and forth. More completion. The model handles messy instructions better and requires fewer prompts to get usable output. If you are on Pro, Business, or Enterprise, you get access to GPT-5.5 Pro which is the more capable variant tuned for serious agent work.

If you build with OpenAI APIs, pricing is key. Standard GPT-5.5 at $5 input and $30 output per million tokens is aggressive pricing against Claude. If your use case does not need the extra reasoning depth of GPT-5.5 Pro, you can probably cut your model costs significantly by switching over. If you do need Pro, the math gets harder because the Pro pricing is above Claude Opus.

If you run an agent or workflow stack, this changes the architecture. Brockman said you can probably collapse three or four GPT-5.4 calls into one or two GPT-5.5 calls for the same task. That is fewer tokens, less glue code, and more reliability per call. The upfront per-token cost is higher but the total task cost often comes out lower.

If you are watching the AI race, today is the inflection point everyone has been waiting for. OpenAI either reasserts dominance with this launch or confirms that Anthropic is the new technical leader. The benchmarks will tell the story over the next two weeks. Enterprise contracts will tell the real story over the next quarter.

OpenAI just shipped GPT-5.5. First fully retrained base model since GPT-4.5. Six weeks after GPT-5.4. Live today for Plus, Pro, Business, and Enterprise across ChatGPT and Codex. 82.7% on Terminal-Bench 2.0. 84.9% on GDPval. State of the art on the Artificial Analysis Coding Index at roughly half the cost of rivals. Greg Brockman called it a new class of intelligence built to finish tasks without human babysitting. Sam Altman's super app vision now has the model that makes it real. OpenAI has been quiet for months while Anthropic shipped product after product. That quiet just ended. The agent era started today and OpenAI is determined to own it. Now we see if the industry lets them.

Quick Links:

GPT 5.5 Release 👉 Here

Stay building. 🤖