
Microsoft releases Aion 1.0 Instruct weights, its 2.3B Windows model
The MIT-licensed small model handles on-device text tasks, and Microsoft says an optimized version becomes the built-in Windows model by January 2027.
762 stories

The MIT-licensed small model handles on-device text tasks, and Microsoft says an optimized version becomes the built-in Windows model by January 2027.
Group revenue rose 10% to RMB271.5 billion, but heavy spending on AI infrastructure pushed free cash flow to a RMB34.5 billion outflow.

Neon costs $3 and $15 per million tokens, more than Argon until November 30 and less once Argon's price doubles to $4 and $20 on December 1.

The Gemini icon shows on the right edge of the map only during navigation. Here is what you need, what to ask, and what to check when it's missing.

Hydrogen plans a task from a camera view, writes the control code and fixes it until the job is done, and a new open-source Antigravity plugin sets up the arm and the simulator.

Anthropic's November 12 plugin drafts quotes, 8Ds and production reports from NetSuite, Infor, Microsoft 365 and Fusion, with no price beyond a paid Claude plan.

OpenAI's beta dashboard lets owners verify their ChatGPT listing, add product feeds and see 28 days of mentions and clicks, with no fee listed.

Free users get Gemini 4 Hydrogen. Neon needs Google AI Pro at $19.99 a month in the Gemini app and AI Studio, or the paid Gemini API at $3 in and $15 out per million tokens.
CEO Lip-Bu Tan announced the deal in Austin without naming the customer; Microsoft is the first named buyer of Intel's EMIB packaging.

Versions 1.26.1 and 1.26.2 were live for about an hour on November 17 and took Azure, GitHub and CI credentials; 1.26.0 is clean.

Both Ultra tiers now come with YouTube Premium Family, alongside 20 TB or 30 TB of storage and 5 or 20 times the Gemini limits of AI Pro.

DeepSeek V4.1-Flash is cheaper per token and leads most shared coding tests, while Z.ai's GLM-5.3 family offers a coding subscription from $18 a month.

Founded in 2021 by seven co-founders, the public benefit corporation took in $33.2 billion in the first nine months of 2026 and priced its IPO at $83 a share.

The chat app is free. The API runs on a prepaid balance paid by PayPal, bank card, Alipay or WeChat Pay, with Flash from $0.15 per million input tokens.

LoRA freezes a model and trains two thin matrices per layer, cutting GPT-3's fine-tuning checkpoint from 350 GB to 35 MB. QLoRA and a new AWS test show its reach and its limits.
Musk says pre-training is done and the model is in reinforcement learning, but as of November 19 Grok 4.8 has no date, price or API listing, only trials.
The verified MCP connectors are in public preview on every Copilot plan, from the app, the CLI, VS Code and github.com.

The SGLang team's first model is Qwen3.8-27B trained further on long agent tasks, and it doubles the base model's Terminal Bench 4.1 score.

The company formerly called Pure Storage raised its fiscal 2027 forecast to $5.2 billion, and its shares rose about 5% after hours.

The 27B Clairic and 9B Clairic-flash pull structured fields from text, images and PDFs, and cost $0.10 per million input tokens on Workers AI until December 1.

Muse runs on an isolated cloud computer and asks before sensitive actions, but Reuters found it read Signal messages through phone permissions people granted at setup.

Google's tables put Neon ahead on 16 of 19 rows, Artificial Analysis scores it 61, and from December 1 it costs less per token than Gemini 4 Carbon.

Text now appears as you talk, per OpenAI's announcement reported by 9to5Mac. How to dictate on desktop and mobile, how it differs from voice mode, and what to check when it fails.

Post-training turns a next-word predictor into a useful assistant, and this week Palo Alto Networks agreed to pay $4.5 billion in stock for a company that does it.