Thursday marked the launch of OpenAI’s GPT-5.5, positioned as an advanced model for agentic computer operations. Capable of writing and debugging code, browsing the web, managing spreadsheets, and executing multi-step tasks independently, it minimizes the need for human oversight.
Currently available to Plus, Pro, Business, and Enterprise subscribers on platforms like ChatGPT and Codex, OpenAI describes GPT-5.5 as its most intuitive model yet. The update is aimed at revolutionizing computer-based work, excelling in agentic coding, knowledge tasks, and early-stage scientific research by reasoning across contexts and executing actions over time.
“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 announced. The model promises to redefine intelligence for real-world tasks and agent-powered applications by understanding complex goals, utilizing tools effectively, verifying its output, and completing more assignments.
On Terminal-Bench 2.0, which evaluates command-line task management involving planning and iterative tool use, GPT-5.5 achieves an 82.7% score—surpassing Claude Opus 4.7 at 69.4%, and Gemini 3.1 Pro at 68.5%. In the GDPval benchmark, testing knowledge work across 44 professions from finance to product management, it matches or surpasses industry experts in 84.9% of cases.
Its coding prowess is also noteworthy. On Expert-SWE, a long-duration internal coding task with an average human completion time of 20 hours, GPT-5.5 outpaces its predecessor, GPT-5.4. In SWE-Bench Pro, assessing real-world GitHub issue resolution, it scores 58.6%, compared to Claude Opus 4.7’s 64.3%. OpenAI attributes this variance to potential memorization issues reported by Anthropic.
This release occurs amidst a fast-evolving market for agentic AI. GPT-5.4 succeeded GPT-5.3 within two days, while Xiaomi transitioned from MiMo-V2-Proto to MiMo 2.5 Pro with full multimodal features in five weeks. The interval between GPT-5.4 and GPT-5.5 spanned seven weeks, reflecting the industry’s rapid pace.
For everyday users not subscribed beyond free access, this model remains out of reach; it is exclusive to Plus subscribers at $20/month and higher tiers. Our attempts to test GPT-5.5 under a Pro account were unsuccessful due to immediate unavailability.
The more significant impact lies in Codex, OpenAI’s agentic coding environment, where the model exhibits superior capabilities. “It genuinely feels like I’m working with a higher intelligence, and there’s almost a sense of respect,” remarked Pietro Schirano, CEO of MagicPath.
GPT-5.5 Pro, tailored for demanding tasks requiring high accuracy, is being phased into Pro, Business, and Enterprise accounts on ChatGPT. It scores 90.1% on BrowseComp, which tests web-based information retrieval skills, outperforming Gemini 3.1 Pro at 85.9%. The model also leads in the Artificial Analysis Index as the most intelligent average model, utilizing tokens more efficiently for superior outcomes.
However, the pricing structure may surprise some: OpenAI’s API will soon charge $5 per million input tokens and $30 per million output tokens. GPT-5.5 Pro rates at $30 per million input tokens and $180 per million output tokens. These figures exceed those of GPT-5.4 ($2.50 and $15.00, respectively) while maintaining the same pricing for GPT-5.5 Pro as its predecessor.
OpenAI CEO Sam Altman contends that token efficiency gains offset these costs, allowing GPT-5.5 to perform Codex tasks with fewer tokens, thus reducing overall expenses despite higher per-token rates. For comparison, Xiaomi MiMo v2.5 Pro charges $1 and $3 per million input/output tokens; Minimax M2.7 is priced at $0.30 and $1.20; Kimi K2.5 requires $0.44 and $2.00.