Frontier AI is being commoditized overnight: Grok 4.6 and DeepSeek V4-Pro both launched with steep price cuts and long-context agentic capabilities, while Claude's browser sessions now sync across devices.

• xAI launched Grok 4.6 at $2/$6 per million input/output tokens (~60% cheaper than GPT-5.6), with 500K context, function calling, and strong performance on long-running agentic and codebase tasks.
• DeepSeek V4-Pro hit general availability with 1M token context, 384K max output, adjustable reasoning effort, native OpenAI Responses API support, and 50% off-peak pricing for production workloads.
• Anthropic's Claude Chrome extension now syncs as a full Cowork session across desktop, web, and mobile, letting Claude take browser actions (clicking, form-filling) with continuity across devices - but prompt injection risk remains a live concern.
• A new tool can strip invisible watermarks from Claude, Gemini, and OpenAI outputs, raising content-provenance and IP-tracking questions for anyone shipping AI-generated creative work.
• Pathway's 150M parameter model solved ARC-AGI tasks at $0.0007 each, setting a new cost-efficiency benchmark that suggests small specialized models can rival larger ones on narrow reasoning tasks.
• New research shows four specific design choices can degrade long-context performance by up to 47%, a practical warning for anyone building RAG or agent pipelines on these new large-context models.

Google announces Gemini 3.7 Flash just three weeks after previous release

OpenAI brings ChatGPT and the Codex coding agent natively to Linux, while researchers show every major model's 'encrypted' reasoning can be decoded by a cheaper model with no encryption broken.

• Researchers showed encrypted reasoning blobs from Claude Opus, GPT, and Gemini are portable across models — feed one to a cheaper model like Haiku and it reads the hidden reasoning aloud, no decryption needed.
• A scan of ~7,000 public Claude Code/Codex session traces found 62 exposed API keys, 33 emails, and 33 passwords hidden inside these 'encrypted' reasoning chunks.
• OpenAI shipped a native ChatGPT desktop app for Linux (Ubuntu, Debian, Fedora) bundling ChatGPT, ChatGPT Work, and Codex, giving the coding agent direct access to local repos and files.
• Unsloth released an open-source desktop app for local model training/inference on Mac, Windows, and Linux — 2x faster training, 70% less GPU memory, connects to Claude Code/Codex as a local API replacement.
• OpenAI added auto-sync of projects, chats, and plugins into Codex.
• Google's open diffusion model hit 1,500 tokens/sec on a single GPU; a new inverse model can reconstruct exact LLM prompts from public outputs alone.

Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage Buckets

OpenAI embeds itself inside IBM Consulting while Alibaba and xAI quietly ship stronger creative-generation models.

• OpenAI-IBM deal creates an 'OpenAI Practice' inside IBM Consulting with thousands of certified consultants to deploy GPT-5.6, Codex, and ChatGPT Work into legacy enterprise systems (finance, HR, security via Project Daybreak).
• Google's Pixel 11 pushes an 'agentic phone' vision — sandboxed multi-step task agents, Rambler voice keyboard, and Magic Capture — while deliberately avoiding the 'AI' label in marketing.
• Mozilla's new CEO is positioning Firefox as model-agnostic (user picks their own AI model in Smart Window), explicitly betting on open-weight ecosystems over building a frontier model.
• Alibaba shipped Wan-Animate-2, a high-fidelity real-time character animation model — directly relevant for creative/video pipelines.
• xAI released Grok 4.6 with improved long-running agent performance and more capable interactive/visual generation.
• Anthropic launched Claude Cowork in Chrome's side panel for cross-tab browser agent work; OpenAI added sync for ChatGPT Work/Codex projects, chats, and plugins.

Claude's new Scarlet Letter watermark is invisible—for now

DeepSeek V4 Pro 0813 (on OpenRouter)

• <p><strong><a href="https://openrouter.ai/deepseek/deepseek-v4-pro-0813">DeepSeek V4 Pro 0813 (on OpenRouter)</a></strong></p> The latest DeepSeek Pro model is now available, via API only. I had to link to OpenRouter because DeepSeek don't have any obvious announcement page for their new model.</p>

LFM2.5-VL-3B for Better and Faster Vision Capabilities for the Edge

Platforms from Anthropic to Spotify to Substack are rolling out watermarking and AI-detection tools just as new data shows humans can barely spot AI content anymore.

• Anthropic now watermarks all Claude-generated text and files (C2PA standard) for models released after Aug 2, to comply with the EU AI Act
• Spotify launched 'AI Personas' badges to flag non-human artists and will exclude them from algorithmic recommendations unless users follow them
• Suno added watermarking/fingerprinting tech and tightened its download policy to fight spam and fraud
• Substack integrated the Pangram AI-detection tool so readers can see what percentage of a post was AI-written
• A Deezer study found 97% of listeners can't tell AI-generated music from human-made tracks; ~50% of Deep View readers miss AI images in its own daily game
• Separately: Oumi launched a platform letting enterprises build/deploy/retrain custom AI models in hours instead of months, and Databricks acquired Electric (PGlite) to let coding agents run locally on phones/tablets

llm-gemini 0.33

• <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-gemini/releases/tag/0.33">llm-gemini 0.33</a></p> <p>It's been a while since the last <code>llm-gemini</code> release. This version of the plugin adds support for today's <a href="https://blog.google/innovation-and-ai/models-and-res

What We Learned by Reproducing 2,200 papers from ICML

The web’s newest weapon against AI scrapers is a font

Quoting Florian Herrengt

• <blockquote cite="https://blog.florianherrengt.com/ai-removing-middle-class-software-engineering.html"><p>But then users start to report a weird bug. It's the 4th time your team has been trying to fix it. I mean... asking AI to fix it. Unfortunately, it seems like not even Fable can figure it out.</

Twitch content has trained Amazon AI for years, but users can opt out now