English translation is being prepared.
Matched fieldsTitle、Summary、Body
Matched fieldsTitle、Summary、Body
Matched fieldsTitle、Summary、Body
Matched fieldsTitle、Summary、Body
Matched fieldsTitle、Summary、Body
Matched fieldsSummary、Body

Matched fieldsTitle、Summary、Body

Matched fieldsTitle、Summary、Body

Matched fieldsTitle、Summary
Microsoft has released the image editing model MAI Image 2.7 Preview, now available on Arena and MAI Playground. The model ranks 2nd on Artificial Analysis's AA-Image-Editing v2.1 leaderboard and 3rd on Image Editing Arena, a 45-point improvement over 2.6. Meanwhile, MAI-Image-2.6 and MAI-Image-2.6-Flash are now generally available in Microsoft Foundry.
Matched fieldsSummary、Body
Matched fieldsSummary、Body
Matched fieldsSummary
Matched fieldsSummary
Matched fieldsSummary、Body
Matched fieldsSummary、Body
Meta releases Muse Spark Open, calling it the most capable US open-weight model to date, with 816B total parameters, 42B activated parameters, a 1 million token context window, and aimed at coding and agent tasks. The model was adapted for Mac Studio in partnership with Apple and can run on two 512GB M5 Ultra Mac Studios via the MLX framework; the 512GB Mac Studio is not yet on sale and is expected to launch later this month. Model weights can be downloaded from Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models, and can also be used via the APIs of Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI; downloads are free under a license permitting limited commercial use.
r Mac Studio in partnership with Apple and can run on two 512GB M5 Ultra Mac Studios via the MLX framework; the 512GB Mac Studio is not yet on sale and is expected to launch later this month. Model weights can be downloaded from Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models, and can also be used via the APIs of Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI; downloads are free under a license permitting limited commercial use.
Matched fieldsSummary、Body
Matched fieldsSummary、Body
Matched fieldsSummary、Body
On October 9, Microsoft launched Microsoft-Decision-1, which does not generate text but instead performs a single forward-pass scoring of preset answer options and returns a calibrated probability for each option, for use in routing, classification, prioritization, and verification. The model is post-trained on Alibaba's Qwen3.5-9B, is in public beta on Microsoft Foundry, and is available on OpenRouter under the identifier microsoft/microsoft-decision-1, with weights not distributed externally. Microsoft says it achieved 83.5% accuracy across 36 benchmarks and about 150,000 held-out questions, and plans to later migrate the base to MAI and OpenAI models; the OpenRouter listing shows a price of $0.042 per million input tokens, with output free.
On October 9, Microsoft launched Microsoft-Decision-1, which does not generate text but instead performs a single forward-pass scoring of preset answer options and returns a calibrated probability for each option, for use in routing, classification, prioritization, and verification. The model is post-trained on Alibaba's Qwen3.5-9B, is in public beta on Microsoft Foundry, and is available on OpenRouter under the identifier microsoft/microsoft-decision-1, with weights not distributed externally.
Matched fieldsSummary
Matched fieldsSummary、Body
Matched fieldsSummary、Body