OpenAI Introduced dots, always-on agents in ChatGPT, powered by GPT-6 Astra.

They'll look familiar if you've used OpenClaw, Muse, Instinct, or Grok Bots.

Your Dot sits in the redesigned ChatGPT sidebar right below the “New chat” button. It has its own cloud computer that lets it work while your laptop is closed. You can text it any time or call it on the phone.

Meta presents Context Language Models

Current harnesses have a very rigid setup in how the LLM can manage its own context.

Researchers show that you can let LLMs edit the context directly - enabling better long-horizon memory, FLOP efficient adaption, & more.

Don't fear compaction anymore.

CLM treats context as an editable file rather than an append-only conversation.

+65% score with the same compute on a 24-hour multi-repository agent-swarm task.

GitHub.

Meet Gemini 4 Argon

It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense.

With a 1M token output limit, Argon adds a deeper level of reasoning to tackle long, multi-step problems in one go.

Argon is priced at $2 in and $10 out during introductory pricing!

DeepSeek released Harness v0.2 preview, with a desktop app available for macOS and Windows.

You can install it here and follow the instructions to get started.

Anthropic just published its 𝗳𝘂𝗹𝗹 𝗴𝗲𝘁𝘁𝗶𝗻𝗴-𝘀𝘁𝗮𝗿𝘁𝗲𝗱 𝗴𝘂𝗶𝗱𝗲 for Claude Code mods

A mod is a small feature you add to Claude Code. The guide's three mods take the things you'd normally stop to ask it and 𝗽𝘂𝘁 𝘁𝗵𝗲𝗺 𝗿𝗶𝗴𝗵𝘁 𝗼𝗻 𝘀𝗰𝗿𝗲𝗲𝗻: how full the context is, what this command is about to delete, and what it just changed.

Keep mods to yourself, or submit them to the Claude directory if others might find them useful. Mods run with the same access to your machine as Claude Code itself, so make sure to only install mods from sources you trust.

Blog post.

Apple introduced LoopCD

Heard the rumors that frontier models like GPT-6 Astra and Gemini 4 use Looped Transformers?

Apple presented a new inference-time method that makes pre-trained looped models better for (almost) free.

No training. No extra model.

Google research introduced a next-generation Federated Learning system.

By using Trusted Execution Environments (TEEs), it delivers verifiable differential privacy while moving computation server-side to cut training times.

Meet Promethee, an app that turns your work into a multiplayer game, to perform.

Random German lab appears from nowhere and drops SOTA open-weight model Kolibri-1

78B parameters, 3.46B active. Up to 1M tokens of context, under Apache 2.0.

Beats Qwen, Mistral, Nematron models with similar active parameter count.

Paper.

Meta introduced Muse Gadgets, an open source ESP32 firmware and Linux sdk so that you can make hardware devices that work with Muse.

Grab an API token from gadgets.muse.ai and point your favorite coding agent at the github repo to build your own peripherals for Muse.

Hugging Face turned Claude Code, Codex, Hermes, Pi, and other coding harnesses into RL environments

No changes to the harnesses, no changes to the training code.

Any open model, any task set, fully open source

Same model, same weights: 62% under Mini-SWE-Agent, 33% under Claude Code.

But training inside a real harness normally means reimplementing it as an environment, so most models get trained in a scaffold nobody actually ships.

Google Introduced AlphaProtein Novo, a pipeline for de novo enzyme design

Researchers developed AP Novo and used it to get SoTA "novel-scaffold" activity on 2 benchmark reactions and create enzymes to synthesize the pharmaceutical motif piperidine and degrade the environmental toxin DEHP.

Preprint
Code

Anthropic removes Cowork's local option for Pro/Max tomorrow.

New tasks run in the cloud. Existing local tasks stay on your computer.

Anthropic says your sessions and files "live with your Claude account and go where you go."

Reflection introduced Beam: an agentic open model with 501B total parameters and 23B active

Full weights release this month.

Mistral launched a preview of new model, Mistral Large 4 (ML4), aka le Chonk

ML4 is a 1T-parameter model with 49B active parameters, trained natively with multimodal capabilities.

Open weights release end of October.

ML4 was trained on 3,800 NVIDIA Grace Blackwell GPUs in own European datacenters - cluster in Bruyères-le-Châtel built.

Vitalik Buterin: AI’ll become the new UI and traditional on-chain applications may no longer exist in 2 years

Today Ethereum co-founder said at OKX NOW in Singapore that AI will become the new user interface in more and more scenarios.

He shared how a local AI agent wrote a script to update his ENS hash in about five minutes, and said performing complex blockchain operations without a traditional UI will become normal.

This could reduce security risks associated with traditional frontends, while introducing new challenges such as prompt injection, AI safety, and whether agents correctly understand onchain operations.

Also he said that over the next two years, Ethereum transaction costs are expected to fall further, privacy capabilities will improve, and the integration of AI with on-chain applications will accelerate.

He said future on-chain interactions may increasingly be handled by bots, while traditional applications could become less prominent, with AI directly interpreting on-chain offers and filtering transactions on users’ behalf.

Sierra announced Personal Agent Protocol - an open standard

It will help define how personal agents interact with businesses and is open for anyone to implement.

Sui partners with Alibaba Cloud to enable stablecoin-based per-call payments for AI agent services