GPT-5.5 Arrives, OpenAI’s Smartest Model Yet Excels at Coding and Research

OpenAI has released GPT-5.5, a new artificial intelligence model that represents a meaningful jump in capability and autonomy. The model is now rolling out to ChatGPT and Codex users, with API access coming soon.

The headline capability: GPT-5.5 understands what you’re trying to accomplish and handles substantially more of the work itself without requiring you to manually guide every step.

What GPT-5.5 Does Differently

Unlike earlier models that excel at answering specific questions, GPT-5.5 tackles messy, multi-step problems end-to-end. Give it a complicated task and it plans the approach, uses available tools, checks its own work, navigates around obstacles, and keeps going until completion.

The model excels especially in four domains:

Agentic coding. On Terminal-Bench 2.0, a benchmark testing complex command-line workflows, GPT-5.5 achieves 82.7% accuracy, state-of-the-art performance. It handles real GitHub issue resolution at 58.6% on SWE-Bench Pro, solving more tasks in a single pass than previous models.

Beyond benchmarks, early testers report that GPT-5.5 demonstrates conceptual clarity about code architecture. It understands why something is failing, where the fix needs to land, and what else in the codebase gets affected. One senior engineer at NVIDIA said losing access to GPT-5.5 “feels like I’ve had a limb amputated”.

Knowledge work. In spreadsheet generation, document creation, and data analysis, GPT-5.5 outperforms GPT-5.4. It moves naturally through the full loop: finding information, understanding what matters, using tools, checking output, and transforming raw material into useful results.

OpenAI’s own teams are already using it. The Communications team used GPT-5.5 to analyze six months of speaking requests, build a scoring framework, and validate an automated system that now handles low-risk requests automatically. The Finance team processed 24,771 tax forms totaling 71,637 pages, work that was accelerated by two weeks compared to the prior year.

Scientific research. GPT-5.5 shows marked improvement in scientific workflows. On GeneBench (evaluating multi-stage scientific data analysis in genetics), it significantly outperforms GPT-5.4. An internal version helped discover a new mathematical proof about Ramsey numbers, a concrete example of AI contributing original research insights.

Researchers are using GPT-5.5 Pro as an active research partner, critiquing manuscripts, stress-testing arguments, and proposing analyses across multiple passes.

Computer use and vision. On OSWorld-Verified (measuring whether a model can operate real computer environments autonomously), GPT-5.5 reaches 78.7% accuracy. It can see what’s on screen, click, type, navigate interfaces, and move across tools with precision.

Speed Without Sacrificing Power

A critical detail: GPT-5.5 matches GPT-5.4’s per-token speed while delivering substantially higher intelligence. Larger, more capable models are typically slower. OpenAI achieved this through rethinking inference as an integrated system and co-designing the model for NVIDIA GB200 and GB300 hardware.

More importantly, GPT-5.5 uses significantly fewer tokens to complete tasks. On the Artificial Analysis Coding Index, it delivers state-of-the-art intelligence at half the cost of competing frontier coding models. This efficiency matters practically: fewer tokens means faster responses and lower API costs.

Safety Built In

OpenAI released GPT-5.5 with what it calls its “strongest set of safeguards to date”. The model went through full safety and governance processes, including preparedness evaluations, domain-specific testing, and external expert red-teaming.

Cybersecurity capabilities are classified as “High” under OpenAI’s safety framework. The company deployed industry-leading safeguards around higher-risk activities and sensitive requests while creating a “Trusted Access for Cyber” program for verified defenders and critical infrastructure organizations.

Availability and Pricing

In ChatGPT: GPT-5.5 is available to Plus, Pro, Business, and Enterprise users. GPT-5.5 Pro (for harder questions and higher-accuracy work) is available to Pro, Business, and Enterprise users. GPT-5.5 Thinking mode is also available to these tiers.

In Codex: GPT-5.5 is available across Plus, Pro, Business, Enterprise, Edu, and Go plans with a 400K context window. A faster mode (1.5x speed at 2.5x cost) is also available.

API pricing: Coming soon at $5 per 1M input tokens and $30 per 1M output tokens (with a 1M context window). GPT-5.5-pro will be priced at $30 per 1M input and $180 per 1M output tokens. Batch and Flex pricing are available at half standard rates.

Real-World Impact Already Visible

Early testers provide concrete examples of what GPT-5.5 enables:

Cursor (AI coding IDE):

GPT-5.5 is noticeably smarter and more persistent than GPT-5.4, with stronger coding performance and more reliable tool use. It stays on task for significantly longer.

Derya Unutmaz, immunology researcher:

Used GPT-5.5 Pro to analyze a 62-sample, 28,000-gene dataset, producing a detailed research report that would have taken his team months.

Bartosz Naskręcki, mathematics professor:

Built a complete algebraic-geometry visualization app in 11 minutes from a single text prompt.

The Competitive Landscape

On major benchmarks, GPT-5.5 leads or matches competitors:

• GDPval (knowledge work across 44 professions): 84.9% vs Claude Opus 4.7’s 80.3%
• Terminal-Bench 2.0 (complex coding workflows): 82.7% vs GPT-5.4’s 75.1%
• CyberGym (cybersecurity tasks): 81.8% vs GPT-5.4’s 79.0%
• BixBench (bioinformatics): 80.5% vs GPT-5.4’s 74.0%

What This Means Going Forward

GPT-5.5 represents a step toward what OpenAI calls “agentic AI”, models that can plan multi-step work, use tools autonomously, and persist through complexity without human intervention at every stage.

The practical implication: software engineers spend less time managing every detail. Researchers move from question to results faster. Knowledge workers offload routine analysis and document work to AI handling the full workflow.

Whether GPT-5.5 becomes the standard in these domains depends on adoption, pricing, and how well organizations integrate it into real workflows. But the capability threshold has clearly moved higher.