OpenAI's GPT-5.5 Launches With 91.7% Benchmark Score

OpenAI has officially unveiled GPT-5.5, its latest AI model, on April 23, 2026, pushing the boundaries of artificial intelligence in professional and technical applications. Early evaluations show the model scoring an impressive 91.7% on Harvey.ai's BigLaw Bench evaluation suite, edging out GPT-5.4's already strong 91.0%. This marks a new high for OpenAI's generative models and positions GPT-5.5 as a powerful tool for legal and technical industries.

Designed to excel in complex and ambiguous tasks, GPT-5.5 showcases advancements in coding, data analysis, and autonomous task execution. In legal-specific testing conducted by Harvey.ai, the model demonstrated notable improvements in risk assessment, deal management, and litigation analysis. It also achieved perfect scores on 43% of tasks, with 87% of tasks scoring above 0.80—indicating consistent performance across diverse use cases. Legal professionals have praised GPT-5.5's ability to produce outputs with superior organizational structure, intuitive formatting, and grounded citations.

“GPT-5.5 shows improvement in legal reasoning, organizational structure, and audience calibration,” said Niko Grupen, Head of Applied Research at Harvey. “These are the kind of practical gains that move the needle for legal practitioners.”

However, verbosity remains a minor challenge, as some responses were more detailed than necessary, particularly on straightforward queries. Despite this, the model’s overall performance and versatility signify a leap forward in AI capabilities for professional applications.

OpenAI has made GPT-5.5 available to paying subscribers across its ChatGPT and Codex platforms. Users on Plus, Pro, Business, and Enterprise tiers gain immediate access, while a 'GPT-5.5 Pro' version is tailored for heavy computational workloads. This release also comes with updated API pricing, reflecting the enhanced functionality and performance of the new model.

With GPT-5.5, OpenAI is aiming to stay ahead of competitors like Anthropic and Google in the race to dominate the AI space. The model's release underscores OpenAI’s strategy to position its AI as autonomous 'digital workers,' capable of handling tasks with minimal human intervention. Greg Brockman, President of OpenAI, stated that GPT-5.5’s ability to solve problems without explicit step-by-step instructions represents a significant leap in AI autonomy.

For users and developers, GPT-5.5 offers not just incremental improvements but a new layer of functionality that could redefine workflows across industries. As organizations begin integrating the model into their operations, the implications will likely shape market dynamics in both AI and legal tech sectors over the coming months.