OpenAI has officially released GPT-5.5, a generational upgrade that has already contributed to a new mathematical proof for Ramsey numbers. The 'Pro' edition of the model nearly doubled previous performance on the FrontierMath Tier 4 benchmark, scoring 39.6% on problems that typically require days for human experts to solve. The release also includes new 'Workspace Agents' designed for complex, multi-step autonomous workflows.
The AI industry is witnessing a significant shift away from the 'open weights' era. Following Meta's surprise move to keep its most powerful Llama 4 variant, 'Muse Spark,' proprietary, and Anthropic's recent experiments with tiered access for its 'Claude Code' agent, major labs are increasingly moving behind APIs to protect competitive advantages and manage the massive compute costs of agentic reasoning.
Google's new 'Deep Research Max' agent can autonomously conduct up to 160 sequential web searches and consult over 100 sources in a single overnight task to generate comprehensive due diligence reports.
Which benchmark did GPT-5.5 Pro set a new record on this week, testing economic utility across 44 different fields?