OpenAI has unveiled GPT-5.5, its latest flagship large language model, positioning it as a step-change in how artificial intelligence interacts with software systems, workflows and enterprise use cases. 

The release follows months of speculation around a next-generation model and comes at a time when competition with rivals like Anthropic and Google has intensified.

The company is pitching GPT-5.5 not just as an upgrade over GPT-5.4, but as a re-architecture focused on “agentic” capabilities, where the model can independently interpret problems and execute multi-step tasks with minimal prompting.

“It’s definitely our strongest model yet on coding, both measured by benchmarks and based on the feedback that we’ve gotten from trusted partners, as well as our own experience,” said Amelia 'Mia' Glaese, vice president of Research at OpenAI.

Must read: OpenAI unveils GPT-5.5 for complex AI-driven tasks, strengthens ‘super app’ roadmap

How is GPT-5.5 different from earlier models?

OpenAI co-founder and president Greg Brockman emphasised the model’s ability to operate with less human intervention. “What is really special about this model is how much more it can do with less guidance. It’s way more intuitive to use. It can look at an unclear problem and figure out what needs to happen next,” he said.

At the heart of GPT-5.5 is this shift toward autonomy. Unlike earlier systems that relied heavily on structured prompts, the new model is designed to handle complex workflows, ranging from debugging large codebases to navigating across documents and spreadsheets, without step-by-step instructions. 

OpenAI says this makes it particularly effective for coding, enterprise software use and scientific research, areas Brockman described as “very intelligent bottlenecks.”

What is GPT-5.5 Pro and who is it for?

The launch also includes a higher-end variant, GPT-5.5 Pro, which is tailored for use cases demanding greater precision, such as legal analysis, data science and high-stakes enterprise decision-making. While the base model is intended for broad applications, the Pro version focuses on deeper reasoning and structured outputs, supported by latency optimisations for multi-step tasks.

Must read: BT explainer: Why Anthropic’s Claude Mythos could become a major security risk?

OpenAI CEO Sam Altman posted on X that “we want our users to have access to the best technology and for everyone to have equal opportunity.”

How does GPT-5.5 perform against rivals?

Benchmark performance suggests OpenAI has regained ground in the intensifying race for leading frontier models. GPT-5.5 outperforms widely available rivals such as Claude Opus 4.7 and Gemini 3.1 Pro across a majority of tested categories, particularly in areas like agentic computer use, cybersecurity tasks and complex mathematics.

On Terminal-Bench 2.0, a test of a model’s ability to complete tasks in a simulated terminal environment, GPT-5.5 recorded 82.7% accuracy, ahead of Opus 4.7 at 69.4% and narrowly above Claude Mythos Preview at 82.0%.

However, the picture is more nuanced in pure reasoning benchmarks without tool use. GPT-5.5 Pro scored 43.1% on Humanity’s Last Exam, trailing Opus 4.7 at 46.9% and Mythos Preview at 56.8%. This indicates that while OpenAI has pushed ahead in applied, task-oriented intelligence, rivals continue to compete strongly in abstract reasoning.

It is worth noting that Mythos Preview is not commercially available and is restricted to select partners due to cybersecurity concerns, leaving the real market contest among publicly accessible models.

What engineering improvements power the model?

One of the notable advances in GPT-5.5 is its efficiency. OpenAI claims the model delivers higher intelligence without increasing latency compared to GPT-5.4. This has been achieved through hardware-software co-design, with deployments on advanced NVIDIA GB200 and GB300 NVL72 systems.

The company said it used custom optimisation techniques, including algorithms generated by the AI itself, to distribute workloads more effectively across GPU clusters, resulting in over 20% faster token generation speeds.

A new “Thinking” mode further allows the model to internally validate its reasoning before responding, producing more concise outputs in complex scenarios.

Who can access GPT-5.5 and what does it cost?

Despite the performance gains, GPT-5.5 is not yet available via API, with OpenAI stating that “API deployments require different safeguards and we are working closely with partners and customers on the safety and security requirements for serving it at scale.”

For now, access is limited to paid tiers of ChatGPT, including Plus, Pro, Business and Enterprise, with GPT-5.5 Pro reserved for higher-tier subscribers.

Pricing for developers is set to rise sharply once API access is enabled. Input and output costs for GPT-5.5 are roughly double those of GPT-5.4, while GPT-5.5 Pro is priced significantly higher still. 

OpenAI argues that improved “token efficiency” offsets some of the cost increase, as the model requires fewer tokens to complete tasks.

What are the safety and cybersecurity implications?

The release also introduces tighter controls around high-risk capabilities. Under OpenAI’s preparedness framework, GPT-5.5 is classified as “High” risk for both biological and cybersecurity domains.

To manage this, the company has implemented stricter safeguards for general users while offering a “cyber-permissive” access programme for verified security professionals. This allows authorised users working on critical infrastructure to use advanced capabilities with fewer restrictions.

The approach reflects a broader industry tension that models are increasingly powerful in identifying vulnerabilities and strengthening systems, but also carry the risk of misuse.