AI news this week

Tuesday, October 13

The one thing. Google shipped Gemini 4 Argon globally on Oct 13 at an introductory $2/$10 per million tokens. It hit Copilot, Cursor and Vercel's AI Gateway the same day, with Copilot access rolling out gradually. Google GitHub Cursor Vercel

• Models Gemini 4 Argon is live in the APISourcedeepmind.google
• Tools Argon is already in your editorSourcesgithub.blog·x.com
• Companies Anthropic outgrew gitSourceanthropic.com
• Models Liquid AI ships an on-device agent modelSourcehuggingface.co
• Models Tune Grok 4.7 effort per requestSourceaws.amazon.com
• Research An embedding model that transcribesSourceeg2.haku.sh
• Companies Character.AI files for Chapter 11Sourcetechcrunch.com

Full Oct 13 issue →

Wednesday, October 14

The one thing. Google shipped Gemini 4 Argon globally on Oct 13, with API access at an introductory $2 per million input tokens and $10 per million output. If you run agentic coding or long multi-step jobs, re-run your evals this week. Google

• Models Argon's API has new plumbingSourcedeepmind.google
• Tools Argon lands in GitHub CopilotSourcegithub.blog
• Companies Anthropic outgrew gitSourceanthropic.com
• Tools Claude Code stops wasting your cacheSourcegithub.com
• Tools Ollama puts decision models in the CLISourcegithub.com
• Companies Anthropic opens in SingaporeSourceanthropic.com
• Funding NVIDIA funds UK agent safety researchSourceblogs.nvidia.com

Full Oct 14 issue →

Thursday, October 15

The one thing. SpaceXSI released Grok Voice 3.0, a speech-to-speech model it says ranks first on Artificial Analysis's Speech-to-Speech Index with a score of 86.9. It costs $0.08 a minute in the Voice Agent API, the same as Grok Voice 2.0, and grok-voice-latest already points to it. xAI

• Products SpaceXSI says Starlink handles 20,000 customer calls a day with Grok VoiceSourcex.ai
• Policy Claude will give US voters state-specific polling informationSourceanthropic.com
• Research Posts say GPT-6 Astra found mistakes in a physics benchmark's answer keySourcex.com
• Companies NVIDIA signs on to the EU's new AI in Science Enablers InitiativeSourceblogs.nvidia.com
• Hardware Swedish firms share a 1,152-GPU supercomputer called Sferical OneSourceblogs.nvidia.com
• Tools Ollama's next release ports its Metal kernels to CUDASourcegithub.com

Full Oct 15 issue →

Friday, October 16

The one thing. NVIDIA released Nemotron 3.5 Super VL on Oct 15. It's an open 120B mixture-of-experts model that activates 12B parameters per token and reads text, images, documents and video. NVIDIA says it gets up to 5x higher agentic throughput on a single GPU than open models its size. NVIDIA

• Safety Anthropic gives six METR staff employee-like accessSourceanthropic.com
• Models OpenAI updates ChatGPT Voice with GPT-Live-1.1Sourcex.com
• Open source Meta releases Muse Spark Open with open weightsSourcex.com
• Open source Ollama adds a smaller Nemotron that fits a 24 GB GPUSourcegithub.com
• Tools Codex CLI 0.163.0 lets you steer a reply mid-streamSourcegithub.com
• Products Copilot in Slack and Teams works across repositoriesSourcegithub.blog
• Research Stanford researchers let agents reason silently between tool callsSourcehuggingface.co

Full Oct 16 issue →