OpenAI Releases GPT-5.5: Faster, Smarter—And Pricier

OpenAI introduced GPT-5.5 on Thursday, pitching it basically as a design targeted at agentic computer system usage. It composes and debugs code, searches the web, completes spreadsheets, and keeps resolving multi-step jobs without requiring a human to babysit every relocation.

The release is currently presenting today to Plus, Pro, Organization, and Business customers throughout ChatGPT and Codex, OpenAI stated.

” We’re launching GPT‑5.5, our most intelligent and most intuitive-to-use design yet, and the next action towards a brand-new method of getting work done on a computer system,” OpenAI stated in a statement. “The gains are particularly strong in agentic coding, computer system usage, understanding work, and early clinical research study– locations where development depends upon thinking throughout context and acting gradually.

The huge heading from OpenAI: GPT-5.5 is measurably smarter than its predecessor, GPT-5.4– and it’s not slower. Matching GPT-5.4’s per-token latency in real-world serving while striking greater ratings throughout criteria is the sort of effectiveness enhancement that typically does not take place. Larger designs tend to be slower when running under the very same hardware.

On Terminal-Bench 2.0, which evaluates how well a design deals with complicated command-line workflows that need preparation and iterative tool usage, GPT-5.5 ratings 82.7%. Claude Opus 4.7 lands at 69.4%, while Gemini 3.1 Pro sits at 68.5%. That’s not a limited lead.

On GDPval, a benchmark screening understanding work throughout 44 genuine professions– from financing to legal research study to item management– GPT-5.5 matches or beats market experts in 84.9% of contrasts.

It’s likewise a respectable coder, as anticipated. On Expert-SWE, an internal criteria for long-horizon coding jobs with a mean approximated human conclusion time of 20 hours, GPT-5.5 outshines GPT-5.4. On SWE-Bench Pro, which grades real-world GitHub concern resolution, it reaches 58.6%. Claude Opus 4.7 ratings greater at 64.3%, however OpenAI declares it might be due to the fact that “Anthropic reported indications of memorization on a subset of issues”

This launch lands in a market that is moving quickly because the boom of agentic AI. GPT-5.4 showed up simply 2 days after GPT-5.3, while Xiaomi went from MiMo-V2-Pro to MiMo 2.5 Pro– with complete multimodal abilities– in approximately 5 weeks. The space in between GPT-5.4 and GPT-5.5 had to do with 7 weeks. That’s the pace now.

However will this design make a distinction for daily users who are not constantly coding the next huge thing? If you’re on a totally free tier, no: GPT-5.5 isn’t pertaining to totally free users. If you’re spending for Plus at $20/month, it presents today. We attempted evaluating it under our Pro account, however the design was not instantly readily available.

The larger offer is most likely what GPT-5.5 does inside Codex– OpenAI’s agentic coding environment– where it is shown to be more effective. “It truly seems like I’m dealing with a greater intelligence, and there’s nearly a sense of regard,” Pietro Schirano, CEO of MagicPath, stated in a quote shared by OpenAI.

GPT-5.5 Pro, developed for more difficult, higher-accuracy work, is presenting individually to Pro, Organization, and Business users in ChatGPT. On BrowseComp, which evaluates a design’s capability to locate hard-to-find details throughout the web, GPT-5.5 Pro ratings 90.1%, ahead of Gemini 3.1 Pro at 85.9%.

The design is likewise the most smart usually based upon the Artificial Analysis Index. GPT 5.5 reports a more effective and beneficial usage of tokens, producing much better lead to basic.

The rates, nevertheless, might surprise some users. The API will charge $5 per million input tokens and $30 per million output tokens when it releases, which OpenAI states is coming “soon.” GPT-5.5 Pro in the API will cost $30 per million input tokens and $180 per million output tokens.

Those figures are greater than GPT-5.4–$ 2.50 per million tokens of input and $15.00 per million tokens of output– while rates for GPT-5.5 Pro stays the like GPT-5.4 Pro.

That stated, OpenAI CEO Sam Altman argued on X that token effectiveness gains balance out the expense– GPT-5.5 finishes the very same Codex jobs with less tokens, which suggests more affordable runs even at a greater per-token rate.

Simply for contrast, Xiaomi MiMo v2.5 Pro charges $1 and $3 per million tokens of input and output, Minimax M2.7 expenses $0.30 and $1.20 respectively, and Kimi K2.5 needs $0.44 and $2.00 per million tokens.