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

OpenAI launched GPT-5.5 on Thursday, pitching it essentially as a model targeted at agentic computer use. It writes and debugs code, browses the web, fills out spreadsheets, and keeps working through multi-step tasks without needing a human to babysit every move.

The release is already rolling out today to Plus, Pro, Business, and Enterprise subscribers across ChatGPT and Codex, OpenAI said.

“We’re releasing GPT‑5.5, our smartest and most intuitive-to-use model yet, and the next step toward a new way of getting work done on a computer,” OpenAI said in an announcement. “The gains are especially strong in agentic coding, computer use, knowledge work, and early scientific research—areas where progress depends on reasoning across context and taking action over time.

The big headline 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 hitting higher scores across benchmarks is the kind of efficiency improvement that usually doesn’t happen. Bigger models tend to be slower when running under the same hardware.

On Terminal-Bench 2.0, which tests how well a model handles complex command-line workflows that require planning and iterative tool use, GPT-5.5 scores 82.7%. Claude Opus 4.7 lands at 69.4%, while Gemini 3.1 Pro sits at 68.5%. That’s not a marginal lead.

On GDPval, a benchmark testing knowledge work across 44 real occupations—from finance to legal research to product management—GPT-5.5 matches or beats industry professionals in 84.9% of comparisons.

It’s also a pretty good coder, as expected. On Expert-SWE, an internal benchmark for long-horizon coding tasks with a median estimated human completion time of 20 hours, GPT-5.5 outperforms GPT-5.4. On SWE-Bench Pro, which grades real-world GitHub issue resolution, it reaches 58.6%. Claude Opus 4.7 scores higher at 64.3%, but OpenAI claims it may be because “Anthropic reported⁠ signs of memorization on a subset of problems”

This launch lands in a market that is moving rapidly since the boom of agentic AI. GPT-5.4 arrived just two days after GPT-5.3, while Xiaomi went from MiMo-V2-Pro to MiMo 2.5 Pro—with full multimodal capabilities—in roughly five weeks. The gap between GPT-5.4 and GPT-5.5 was about seven weeks. That’s the tempo now.

But will this model make a difference for everyday users who are not always coding the next big thing? If you’re on a free tier, no: GPT-5.5 isn’t coming to free users. If you’re paying for Plus at $20/month, it rolls out today. We tried testing it under our Pro account, but the model was not immediately available.

The bigger deal is probably what GPT-5.5 does inside Codex—OpenAI’s agentic coding environment—where it is proven to be more powerful. “It genuinely feels like I’m working with a higher intelligence, and there’s almost a sense of respect,” Pietro Schirano, CEO of MagicPath, said in a quote shared by OpenAI.

GPT-5.5 Pro, designed for harder, higher-accuracy work, is rolling out separately to Pro, Business, and Enterprise users in ChatGPT. On BrowseComp, which tests a model’s ability to track down hard-to-find information across the web, GPT-5.5 Pro scores 90.1%, ahead of Gemini 3.1 Pro at 85.9%.

The model is also the most intelligent on average based on the Artificial Analysis Index. GPT 5.5 reports a more efficient and useful use of tokens, producing better results in general.

The pricing, however, could shock some users. The API will charge $5 per million input tokens and $30 per million output tokens when it launches, which OpenAI says is coming “very soon.” GPT-5.5 Pro in the API will cost $30 per million input tokens and $180 per million output tokens.

Those figures are higher than GPT-5.4—$2.50 per million tokens of input and $15.00 per million tokens of output—while pricing for GPT-5.5 Pro remains the same as GPT-5.4 Pro.

That said, OpenAI CEO Sam Altman argued on X that token efficiency gains offset the cost—GPT-5.5 completes the same Codex tasks with fewer tokens, which means cheaper runs even at a higher per-token rate.

Just for comparison, Xiaomi MiMo v2.5 Pro charges $1 and $3 per million tokens of input and output, Minimax M2.7 costs $0.30 and $1.20 respectively, and Kimi K2.5 requires $0.44 and $2.00 per million tokens.