SAN FRANCISCO — The artificial intelligence industry moved into a heightened competitive posture on Saturday as detailed benchmark results from OpenAI's newly released GPT-5.5 circulated among researchers and enterprise customers, prompting public responses from Google DeepMind and Anthropic about the status of their own next-generation systems.

OpenAI released GPT-5.5 on Thursday, billing it as its most capable model to date for coding, scientific reasoning, and general-purpose work. By Saturday, independent evaluations posted to Hugging Face and the LMSYS Chatbot Arena leaderboard were drawing significant attention, with GPT-5.5 posting strong scores on established coding and mathematics benchmarks including HumanEval and MATH-500, fuelling discussion across developer communities on GitHub and X.

Google DeepMind responded by confirming that Gemini 2.0 Ultra, its flagship enterprise model, remains on track for a broader rollout in early May, with a spokesperson stating the company was 'confident in the capabilities of our upcoming release.' Anthropic, meanwhile, acknowledged in a post on its research blog that Claude 4 — previously expected in late spring — was entering final evaluation stages, a signal widely interpreted by analysts as a competitive response to the pressure created by GPT-5.5's arrival.

Enterprise technology buyers are watching the developments closely. Several Fortune 500 companies currently in annual AI contract negotiations told analysts at Morgan Stanley they were pausing decisions pending a fuller assessment of GPT-5.5 and expected competing offers. Microsoft, which distributes OpenAI models through Azure, confirmed Saturday that GPT-5.5 access would be available to Azure OpenAI Service customers beginning the following week, adding commercial weight to the rollout.

The rapid succession of model releases has reignited debate among AI safety researchers about evaluation standards and deployment pace. The Center for AI Safety in San Francisco published a brief Saturday morning urging all major labs to share evaluation methodology transparently, noting that independently reproducible benchmarks remain the only reliable basis for comparing safety-relevant capabilities across systems. OpenAI said it would publish a full technical report for GPT-5.5 within ten days.