This week's AI news highlights major releases from xAI, Alibaba, and DeepSeek, alongside updates to open video models and inference infrastructure.
- xAI released Grok 4.6, scoring 61 on the Intelligence Index with strong agentic results at $2/$6 per 1M tokens.
- Alibaba released Qwen3.8-Max, a 2.4T parameter MoE model with 95B active parameters, available as open weights.
- DeepSeek V4 Pro GA launched with pricing around $0.435/M input and $0.87/M output, undercutting competitors.
- Microsoft introduced MAI-Thinking-1, its first reasoning model built from scratch, now available in Foundry.
- Lightricks' LTX-2.5 added joint video and audio generation to Diffusers, supporting tile rendering and 48 kHz audio.
- vLLM added Azure Blob support for model loading and KV connectors, enabling faster weight loading via Dynamo ModelExpress.
- Unsloth demonstrated shrinking Qwen3.8-2.4T-A95B from 4.9 TB to 397 GB using dynamic 1-bit quantization.