TL;DR: Microsoft announced MAI-Thinking-1, a 35B parameter reasoning model targeting high-end performance at a lower footprint.
Summary: Microsoft announced MAI-Thinking-1, a 35-billion parameter reasoning LLM available to select early partners. While Microsoft claims the model is preferred to Claude 3.5 Sonnet, early benchmark data shows unexpectedly low scores on GPQA and Terminal Bench 2.0. Access is currently limited, making independent performance verification difficult.
Why it matters: It highlights the industry shift toward smaller, more cost-effective reasoning LLMs. Developers should monitor partner feedback and benchmark validations before planning integrations.
Source: @emollick