OpenAI launches GPT-5.5: faster, smarter, more affordable

OpenAI launched GPT-5.5 on Thursday, offering it primarily as a model aimed at proxy 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 take care of each step.

The release has already been rolled out today to Plus, Pro, Business and Enterprise subscribers via ChatGPT and Codex, OpenAI said.

“We’re launching GPT‑5.5, our smartest, most user-friendly model yet, and the next step towards a new way of doing work on the computer,” OpenAI He said In an advertisement. “The gains are particularly strong in agent cryptography, computer use, cognitive work, and early scientific research – areas where progress depends on thinking across context and taking action over time.

The big title from OpenAI: GPT-5.5 is noticeably smarter than its predecessor, GPT-5.4, and it’s no slower. Matching the GPT-5.4 response time for every token in a real-world service while achieving higher scores across benchmarks is the kind of efficiency improvement that doesn’t normally happen. Larger models tend to be slower when running under the same hardware.

In 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 scored 82.7%. The Claude Opus 4.7 scored 69.4%, while the Gemini 3.1 Pro scored 68.5%. This is not marginal progress.

In GDPval, the standardized knowledge test running across 44 real-world professions—from finance to legal research to product management—the GPT-5.5 matches or beats industry professionals in 84.9% of comparisons.

He’s also a very good programmer, as expected. In Expert-SWE, an internal benchmark for long-running coding tasks with an estimated average human completion time of 20 hours, GPT-5.5 outperforms GPT-5.4. In SWE-Bench Pro, which ranks the accuracy of a real-world GitHub issue, it reaches 58.6%. Cloud Opus 4.7 received a higher score of 64.3%, but OpenAI claims this may be due to “reporting of human save marks on a subset of problems.”

This launch falls into a market that has been moving rapidly since the boom in agentive AI. GPT-5.4 arrived just two days after GPT-5.3While Xiaomi went from Memo V2 Pro to Memo 2.5 Pro– With full multimedia capabilities – in approximately five weeks. The gap between GPT-5.4 and GPT-5.5 was about seven weeks. This is the rhythm now.

But will this model make a difference for everyday users who aren’t always coding the next big thing? If you’re in a free tier, no: GPT-5.5 will not be available to free users. If you pay for Plus at $20 a month, it’s launching today. We tried testing it within our Pro account, but the template wasn’t immediately available.

Perhaps the biggest deal is what GPT-5.5 does inside Codex – OpenAI’s proxy encryption environment – where it has been proven to be more robust. “I really feel like I’m working with a higher intelligence, and there’s almost a feeling of respect,” MagicPath CEO Pietro Schirano said in a quote shared by OpenAI.

GPT-5.5 Pro, designed for more demanding, higher-resolution work, is being released separately for ChatGPT’s Pro, Business, and Enterprise users. In BrowseComp, which tests the model’s ability to track down hard-to-find information across the web, GPT-5.5 Pro scored 90.1%, ahead of Gemini 3.1 Pro with 85.9%.

The model is also the most intelligent on average based on the synthetic analysis index. GPT 5.5 indicates a more efficient and useful use of codes, resulting in better results overall.

However, the price may shock some users. The API will charge $5 per million input tokens and $30 per million output tokens at launch, which OpenAI says will come “very soon.” GPT-5.5 Pro API will cost $30 per million input codes and $180 per million output codes.

These numbers are higher than GPT-5.4 — $2.50 per million input tokens and $15.00 per million output tokens — while the price of GPT-5.5 Pro remains the same as GPT-5.4 Pro.

However, OpenAI CEO Sam Altman said Argue over X The token efficiency gains more than offset the cost – GPT-5.5 completes the same Codex tasks with fewer tokens, which means cheaper runs at even a higher rate per token.

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