OpenAI Unleashes GPT-5.5: Smarter, Faster, and Cheaper
OpenAI has officially launched GPT-5.5, their latest frontier model designed to be smarter and more intuitive. This new model is integrated into services like CodeX and ChatGPT Pro, promising a new way to get work done on computers.
Early testers have been using GPT-5.5 for weeks, finding it significantly improved over its predecessor, GPT-5.4. A key upgrade is its enhanced ‘personality,’ making it less rigid and more conversational, addressing feedback that GPT-5.4 felt soulless.
Focus on Coding and Enterprise
The real power of GPT-5.5 shines in areas like agentic coding, computer use, knowledge work, and scientific research. This focus aligns with market trends, especially following Anthropic’s rapid growth in the enterprise coding sector.
OpenAI’s strategy appears to be a self-improving flywheel: build a strong coding model, sell it to businesses, gather valuable coding data, and use that data to train even better models. This cycle allows for continuous improvement and faster development.
Performance Without Compromise
Despite being a larger and more capable model, GPT-5.5 maintains the same token latency as GPT-5.4. This means users experience similar speeds while getting much more intelligent results.
GPT-5.5 is also more token-efficient, meaning it uses fewer tokens to complete the same tasks. While the cost per token is higher, the overall cost for completing tasks can be lower due to this efficiency.
Improved User Experience
One noticeable improvement is GPT-5.5’s ability to explain complex changes concisely. Instead of lengthy, essay-like explanations, it provides clear, short answers, making it easier for users to understand code modifications.
This more direct communication style is part of the enhanced ‘personality’ of GPT-5.5. It aims to provide users with exactly what they need, without unnecessary jargon or lengthy explanations.
Enhanced Safeguards
OpenAI emphasizes that GPT-5.5 comes with its strongest set of safeguards yet. These are designed to prevent misuse while ensuring access for beneficial applications.
The model is rolling out to various user tiers, including Plus, Pro, Business, and Enterprise users across ChatGPT and CodeX. GPT-5.5 Pro is specifically available for Pro, Business, and Enterprise users within ChatGPT.
Real-World Performance: Benchmarks and Box AI
Box, a cloud content management company, provided benchmark data comparing GPT-5.5 to GPT-5.4. In Box AI’s complex work evaluation, GPT-5.5 showed a significant jump in accuracy, from 67% to 77%.
Industry-specific benchmarks also showed strong gains: financial services saw nearly a 20-point increase, healthcare rose from 61% to 78%, and the public sector improved from 59% to 72%. Media and entertainment saw a 13% jump.
Agentic Coding and Terminal Use
On the Terminal Bench, which tests a model’s ability to operate command-line interfaces, GPT-5.5 showed a 7-point improvement over GPT-5.4. It also significantly outperformed competitor models like Claude Opus 4.7 in this area.
This capability is crucial for agentic usage, allowing AI to interact more effectively with computer systems. The preference for command-line interfaces (CLIs) over slower, error-prone graphical interfaces for AI agents is a key takeaway.
Computer Control and Web Browsing
In computer control benchmarks like OS World Verified, GPT-5.5 showed a 3.7% improvement, matching Claude Opus 4.7. While not a massive leap, it contributes to the model’s overall enhanced capabilities.
For web browsing, GPT-5.5 scored 84.4% on the Browse Comp benchmark, with GPT-5.5 Pro performing exceptionally well in tasks requiring extensive web research. This is a slight improvement over GPT-5.4 Pro and significantly ahead of Claude Opus 4.7.
Advanced Capabilities in Math and Reasoning
GPT-5.5 and GPT-5.5 Pro achieved the top scores on the Frontier Math benchmarks, indicating strong performance in complex mathematical problem-solving. This highlights the model’s advanced reasoning abilities.
The Artificial Analysis Intelligence Index shows GPT-5.5 scoring higher and being more efficient per token than GPT-5.4. This means users get more intelligent output for fewer computational resources, making it cost-effective despite a higher per-token price.
Token Efficiency and Cost Savings
A core focus for GPT-5.5 is token efficiency. While the price per token is higher, the model uses significantly fewer tokens for tasks, leading to a lower overall cost for most use cases. This makes advanced AI more accessible.
For everyday users, this efficiency means getting the same or better results than GPT-5.4 but at a lower effective price. For demanding tasks like frontier science or complex coding, the higher ceiling of GPT-5.5 Pro offers maximum intelligence when needed.
Autonomous Development and Visual Inspection
GPT-5.5 demonstrates a superior ability in visual inspection and iteration within CodeX. It can analyze on-page elements and continuously improve code until a desired outcome is achieved, acting more autonomously.
This enhanced self-correction capability allows GPT-5.5 to complete projects with less human intervention. For example, it can identify and fix issues like misaligned buttons without explicit instructions.
Knowledge Work and Document Generation
The model excels at generating documents, spreadsheets, and presentations. Testers have created coherent, well-designed 60-page documents with ease, showcasing its power in real-world knowledge work.
Companies are already integrating GPT-5.5 into their workflows. For instance, the communications team can analyze extensive data to build risk frameworks and automate low-risk requests, freeing up human resources.
Scientific Research and Future Potential
GPT-5.5 is expected to drive significant improvements in scientific research. Its higher intelligence ceiling allows for tackling more complex problems and exploring new frontiers in discovery.
The model’s ability to understand system logic, identify failure points, and predict the impact of changes sets it apart. This intuitive understanding of codebases allows it to solve problems without needing direct access to live production data.
Availability and Pricing
GPT-5.5 is rolling out to ChatGPT and CodeX users, with GPT-5.5 Pro available for specific tiers. API access for GPT-5.5 is expected soon but not yet available.
The pricing for GPT-5.5 is higher per token ($5 per million input, $30 per million output) compared to GPT-5.4. However, the increased token efficiency is intended to offset these costs for most common applications.
OpenAI’s development of GPT-5.5, particularly its speed and efficiency, was aided by NVIDIA’s GB200 and GB300 NVLink systems. CodeX played a vital role in accelerating the development and testing process.
Source: OpenAI just dropped GPT-5.5… (WOAH) (YouTube)
