OpenAI updated its Microsoft partnership to allow the company to distribute products across all major cloud platforms while maintaining Microsoft as its primary cloud partner. The agreement extends product commitments through 2032 and revenue sharing through 2030.

The shift means OpenAI can now offer models via Google TPU, AWS Trainium, and Bedrock, with Microsoft's license to OpenAI intellectual property becoming non-exclusive. AWS confirmed OpenAI models will arrive on Bedrock within weeks.

The partnership changes effectively remove the previous AGI clause that would have ended Microsoft's exclusive access once artificial general intelligence was achieved.

GPT-5.5 Shows Mixed Performance Gains

Community evaluations show GPT-5.5 achieved 67.1% on WeirdML benchmarks, up from 57.4% for GPT-5.4, but still trailing Anthropic's Opus 4.7 at 76.4% while using fewer tokens.

LMSYS Arena results placed GPT-5.5 at ninth in Code Arena, sixth in Document tasks, and seventh in Text performance.

Meanwhile, GPT-Image-2 continues driving creative applications that demonstrate multimodal AI's potential for general intelligence. Users are generating everything from educational infographics to pop culture content and precise technical diagrams.

The model's integration with Codex allows developers to generate visual assets while coding, creating a closed-loop creative workflow.

Why Multimodal Matters for AGI

The continued success of image generation models like GPT-Image-2, Nano Banana, and Grok Imagine raises questions about resource allocation for labs pursuing AGI. The evidence suggests multimodal capabilities are essential rather than optional.

Text and code generation alone cannot fully demonstrate the "general" aspect of artificial general intelligence. Visual and voice generation capabilities allow AI systems to flex broader reasoning abilities across domains.

The progression from early challenges like "horse-riding astronauts" to current sophisticated visual reasoning demonstrates how image generation has become a key benchmark for general intelligence capabilities.