Published: April 23, 2026
OpenAI has released GPT-5.5, its most capable model to date and the first fully retrained base model since GPT-4.5. GPT-5.5 is a natively omnimodal model that processes text, images, audio, and video in a single unified system, ships with a 1 million token context window, and is designed from the ground up for agentic workflows. The model is available to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with API access rolling out shortly after.
What is GPT-5.5 and what makes it different?
OpenAI describes GPT-5.5 as a model that “understands what you’re trying to do faster and can carry more of the work itself.” Unlike previous GPT models that primarily responded to individual prompts, GPT-5.5 is built to take a sequence of actions, use tools, check its own work, and continue until a task is finished, without requiring the user to re-prompt at every step. This makes it OpenAI’s first model explicitly positioned as an agent rather than a conversational assistant.
GPT-5.5 ships in three variants. The standard GPT-5.5 model handles general-purpose tasks. GPT-5.5 Thinking adds extended chain-of-thought reasoning, which improves performance on math and complex reasoning tasks at the cost of higher latency and token usage. GPT-5.5 Pro delivers the highest accuracy and is aimed at professional and enterprise workflows. The release comes just six weeks after GPT-5.4 shipped, reflecting the accelerating pace of model releases across the industry.
GPT-5.5 benchmarks and performance
GPT-5.5 achieves state-of-the-art results across 14 benchmarks, according to OpenAI. On Terminal-Bench 2.0, which tests complex command-line workflows requiring planning, iteration, and tool coordination, GPT-5.5 scores 82.7%. On OSWorld-Verified, which measures the ability to operate software through graphical interfaces, it reaches 78.7%. On GDPval it scores 84.9%, and on Tau2-bench Telecom it achieves 98.0% without prompt tuning. GPT-5.5 tops the Artificial Analysis Intelligence Index with a score of 60, three points ahead of Claude Opus 4.7 and Gemini 3.1 Pro Preview, both at 57.
On coding-specific evaluations, Expert-SWE reaches 73.1%, up from 68.5% in the previous generation. On MMMU Pro (with tools) it scores 83.2%. On Humanity’s Last Exam (HLE) without tools, GPT-5.5 scores 41.4%, trailing Claude Opus 4.7 at 46.9% and Gemini 3.1 Pro at 44.4%. This suggests that while GPT-5.5 leads on agentic and tool-use tasks, the gap in pure knowledge recall and academic reasoning has not fully closed against its closest competitors.
GPT-5.5 pricing and API availability
GPT-5.5 is priced at $5 per million input tokens and $30 per million output tokens in the API, which is double the cost of GPT-5.4. GPT-5.5 Pro costs $30 per million input tokens and $180 per million output tokens. OpenAI notes that despite the higher per-token cost, GPT-5.5 is more token-efficient than GPT-5.4, delivering better results with fewer tokens for most use cases, which may offset the price increase in practice.
In ChatGPT, GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users. GPT-5.5 Pro is available to Pro, Business, and Enterprise tiers. In Codex, the developer-focused coding environment, GPT-5.5 operates with a 400,000 token context window. API access is expected to follow shortly, with OpenAI noting that API deployments “require different safeguards.”
How does GPT-5.5 compare to Claude Opus 4.7 and Gemini 3.1 Pro?
GPT-5.5 leads on agentic benchmarks, particularly Terminal-Bench 2.0 (82.7%) and OSWorld-Verified (78.7%), where the ability to use tools and complete multi-step tasks is measured directly. It also holds the top position on the Artificial Analysis Intelligence Index at 60 points. Claude Opus 4.7 retains a lead on SWE-Bench Pro, MCP Atlas, Humanity’s Last Exam (no tools), and FinanceAgent v1.1. Gemini 3.1 Pro leads on ARC-AGI-1 and holds slight advantages on GPQA Diamond.
The competitive picture in April 2026 is that no single model dominates all categories. GPT-5.5 is strongest when a task requires autonomous tool use and multi-step execution. Claude Opus 4.7 excels at code generation and knowledge-intensive reasoning. Gemini 3.1 Pro shows particular strength in abstract reasoning benchmarks. For developers and enterprises, the choice increasingly depends on the specific workload rather than overall model ranking.
GPT-5.5 context window and architecture
GPT-5.5 is the first OpenAI API model to ship with a 1 million token context window, a fourfold increase over GPT-5.4’s 256,000 tokens. The architecture is natively omnimodal, meaning text, images, audio, and video are processed end-to-end in a single model rather than through separate pipelines stitched together. This is a continuation of the approach OpenAI introduced with GPT-4o, but at a significantly larger scale.
OpenAI describes GPT-5.5 as the first fully retrained base model since GPT-4.5. Previous updates in the GPT-5 series (5.1 through 5.4) were incremental refinements on the same base. GPT-5.5 represents a new training run, which accounts for the broader gains across benchmarks rather than improvements concentrated in specific areas. The model’s internal codename during development was “Spud.”
The full announcement and technical details are available on OpenAI’s blog at openai.com/index/introducing-gpt-5-5/.
