OpenAI Releases GPT-5.5: Faster, Smarter

On Thursday, OpenAI unveiled GPT-5.5, a next-generation AI model specifically engineered for autonomous computer agents. Designed to operate with minimal human oversight, it excels at writing and debugging code, navigating web interfaces, populating spreadsheets, and executing complex, multi-step workflows independently.

The updated model is now live across ChatGPT and OpenAI’s Codex platform, immediately accessible to subscribers on the Plus, Pro, Business, and Enterprise tiers.

“GPT-5.5 represents our most capable and intuitive model to date, marking a pivotal step toward redefining how computers handle work,” OpenAI stated in its official announcement. “We are seeing particularly strong breakthroughs in autonomous coding, computer interaction, knowledge-intensive tasks, and early-stage scientific research—all areas that demand robust cross-context reasoning and sustained action over time.”

A standout feature of this release is that GPT-5.5 delivers a substantial intelligence leap over GPT-5.4 without compromising speed. Typically, scaling up model parameters increases latency on identical hardware, yet OpenAI has managed to maintain the same per-token latency as its predecessor while achieving superior benchmark scores—a rare efficiency milestone.

When evaluated on Terminal-Bench 2.0, which measures proficiency in managing intricate command-line workflows requiring strategic planning and iterative tool utilization, GPT-5.5 secured an impressive 82.7% accuracy. This significantly outpaces Claude Opus 4.7 at 69.4% and Gemini 3.1 Pro at 68.5%, demonstrating a commanding performance gap.

On GDPval, a comprehensive benchmark assessing knowledge work across 44 real-world professions ranging from finance and legal research to product management, the new model matched or surpassed human experts in 84.9% of direct comparisons.

As anticipated, the model also demonstrates exceptional coding capabilities. On Expert-SWE, an internal benchmark for extended development tasks with a median human completion time of 20 hours, GPT-5.5 outperforms GPT-5.4. It achieved a 58.6% success rate on SWE-Bench Pro, which evaluates real-world GitHub issue resolution. While Claude Opus 4.7 scored slightly higher at 64.3%, OpenAI noted that Anthropic’s results may reflect potential data memorization on certain test cases.

This launch arrives amid an accelerated development cycle in the autonomous AI sector. The industry is moving at a breakneck pace: GPT-5.4 followed GPT-5.3 by just two days, while Xiaomi transitioned from MiMo-V2-Pro to the fully multimodal MiMo 2.5 Pro in roughly five weeks. The seven-week gap between GPT-5.4 and GPT-5.5 underscores the current velocity of AI innovation.

For everyday users who aren’t constantly engineering large-scale projects, the practical impact varies. Free-tier subscribers will not receive access to GPT-5.5. However, Plus plan members paying $20 monthly gain immediate access. During our testing phase on a Pro account, the model was not instantly available, indicating a staggered rollout.

Perhaps more significant is GPT-5.5’s performance within Codex, OpenAI’s dedicated agent coding environment, where it exhibits markedly enhanced capabilities. “Working with it genuinely feels like collaborating with a higher form of intelligence, almost commanding a sense of professional respect,” remarked Pietro Schirano, CEO of MagicPath, in a quote shared by OpenAI.

Additionally, GPT-5.5 Pro, optimized for highly complex and precision-critical tasks, is rolling out separately to Pro, Business, and Enterprise subscribers via ChatGPT. On BrowseComp, a benchmark designed to evaluate a model’s ability to locate obscure information across the web, GPT-5.5 Pro achieved a 90.1% success rate, surpassing Gemini 3.1 Pro’s 85.9%.

According to the Human Analysis Index, the model ranks as the most consistently intelligent on average. GPT-5.5 demonstrates highly optimized token utilization, delivering more efficient processing and consistently superior outputs.

However, the new pricing structure may raise eyebrows. Upon its imminent API release, standard GPT-5.5 will be priced at $5 per million input tokens and $30 per million output tokens. The API version of GPT-5.5 Pro commands a premium rate of $30 per million input tokens and $180 per million output tokens.

These rates represent a notable increase over GPT-5.4, which was priced at $2.50 per million input tokens and $15.00 per million output tokens. Meanwhile, GPT-5.5 Pro maintains parity with the previous GPT-5.4 Pro pricing.