STORY · MODELLER_
Microsoft presents MAI-Thinking-1 with top results on mathematics, Google launches Gemma 4 12B
Microsoft published a technical report for MAI-Thinking-1, a reasoning model that achieves 97% on AIME 2025 and outperforms Claude Sonnet 4.6 in human preference tests, along with Frontier Tuning for efficient adaptation and new models for images and code. In parallel, Google launched Gemma 4 12B, a compact multimodal model under Apache 2.0 license designed for on-device execution with an innovative encoder-free architecture.
WHY IT MATTERS
MAI-Thinking-1 demonstrates significant progress in complex mathematical reasoning without synthetic data dependency, shifting expectations around model transparency and training methods. Gemma 4 12B makes advanced multimodal AI accessible locally, lowering the barrier for integration on end-user devices.
SOURCES
MACHINE-GENERATED SUMMARY This summary is written by machine from the sources below. We sort and explain — but we are a way into the field, not the final word. Check the source when something matters to you.