OpenAI Launches GPT-5.5 — Its Smartest Model Yet That Actually Works Autonomously

OpenAI just dropped GPT-5.5, calling it "the smartest and most intuitive model" they've ever built — and early testers seem to agree. One NVIDIA engineer said losing access to it felt "like being amputated."

The headline capability: GPT-5.5 can take messy, multi-step tasks and handle them end-to-end. It plans, uses tools, verifies its own results, and navigates ambiguity — without needing hand-holding at every step.

Key improvements over GPT-5.4:

• Faster responses with fewer tokens consumed
• Outperforms competing models from Google and Anthropic across benchmarks
• Strongest safety guardrails OpenAI has shipped to date
• Excels at agentic coding, enterprise knowledge work, and scientific research

Two tiers are available: GPT-5.5 Thinking for standard use, and GPT-5.5 Pro for high-precision tasks requiring deeper reasoning.

Availability is rolling out now to ChatGPT Plus, Pro, Business, and Enterprise subscribers. API access is coming soon, priced at $5/M input tokens and $30/M output — double GPT-5.4's rates, but OpenAI argues the efficiency gains offset the cost.

Co-founder Greg Brockman framed GPT-5.5 as a foundation for building a "Super App" — more capable AI that paradoxically uses fewer resources. Simon Willison, who had early access, noted: "I ask it to build things and it builds exactly what I ask for."

The catch? Researcher Ethan Mollick warns GPT-5.5 still exhibits the "jagged frontier" — brilliant at some tasks, unpredictably weak at others. The leap is real, but don't throw away your judgment just yet.