Ashlr Brief.

AWS released general availability of training sessions in SageMaker AI serverless model customization. Builders can now write their own post-training loops, including custom loss functions and multi-turn RL, while SageMaker handles GPU allocation, distributed execution, and fault recovery.

Google and Hugging Face announced native TPU support for PyTorch. Compatibility, warm-up time and cost still need workload-specific checks.

MiniMax says M3.1 Flash can run agent workloads for less. Its posted token prices are low; the useful comparison is what you pay for a completed job, including retries.

Inception launched Mercury 3, a diffusion language model it says runs above 1,000 tokens per second with a 1 million token context window.

GitHub added OpenAI's GPT-6 Sol and Luna to Copilot on September 22, and Codex CLI added both to its model picker on September 23. Access differs by plan.

OpenAI released GPT-6 Sol and Luna with default pricing halved versus GPT-5.6 equivalents. GPT-6 Sol is $2/$10 per million input/output tokens and Luna is $0.10/$0.50, alongside improved prompt caching and more direct response style.

xAI says Grok 4.7 is its most capable model for coding and knowledge work and kept the entry price of Grok 4.6. Access is split across the API, Cursor, Grok Build and a gradual GitHub Copilot rollout.

OpenAI says its GPT-6 Astra configuration for legal work pairs an index of more than 230 million US legal sources with partner plugins. Access starts as invitation-only; the API is not out yet.

Apple says Siri AI began rolling out on September 14 across its OS 27 releases: English only, opt-in, absent from EU iPhone, iPad and Watch, unavailable in mainland China, and capped daily on server-backed features.