Axion 1.0: Open Agent Recipe, Community 35B Weights
Axion 1.0 is an open recipe for an AI agent model. The 35B Apache 2.0 checkpoint is a community reproduction, and its scores sit within 1.8 points of the paper.

Axion 1.0 is an open recipe for an AI agent model. The 35B Apache 2.0 checkpoint is a community reproduction, and its scores sit within 1.8 points of the paper.

OpenAI says its autonomous pipeline validated and patched 948 Linux kernel privilege escalation flaws in a run that cost $1.2 million in API credits.

The Model Context Protocol (MCP) is an open standard for connecting AI apps like Claude and ChatGPT to files, databases and tools. Here is how hosts, clients and servers work, what tools, resources and prompts are, who runs MCP, and what to watch for.

YC Labs and Olam Labs ran swarms of GPT-6.1 Sol and Claude Opus 5.5 agents in Minecraft and Mindustry. Commenters on X and Hacker News joked about the agent bureaucracy, questioned the roster-order hierarchy, and pushed back on the source code headline.

The AI Chrome extensions worth installing as of October 2026, from Claude in Chrome to Tactiq, with what each does and the limits vendors state.

Andy Jassy says shopping with outside AI agents is "still not great." But Amazon blocks Meta's Muse from its store. It should publish terms for outside agents.

Google DeepMind researchers argue smarter AI shifts wages gradually, while AI agents that coordinate well could abruptly reshape firms, jobs and markets.

A new paper trains only per-head attention gates, 46,080 scalars on Llama-3.3-70B, and cuts the best RL-Hammer attack on that model to 1%.

Factory's Droid 1.0 lets models call tools from a short program; Factory reports higher scores than Codex and Claude Code in its Terminal-Bench 4 tests.

LangChain's Managed Deep Agents v0.10, in public beta, adds a remote browser with four tools and a handoff so people sign in without sharing passwords.

Across eight models, higher effort lost tasks that lower effort solved and cost up to 42x as much on easy ones. Starting low and retrying high won.

AI in support helps agents summarize and write, or answers customers itself. What the research says, how Zendesk and Intercom Fin work, and a rollout plan.

Collective raised $100M, a $40M Series B plus a $60M General Catalyst growth facility, to run bookkeeping, payroll and taxes for solo firms with AI.

GitLab Orbit is now GA: a permission-aware graph of code, merge requests and pipelines that Claude Code, Codex and Cursor query via MCP or the CLI.

Deepgram is acquiring Daily, maker of the open-source Pipecat voice agent framework. Pipecat stays BSD-licensed, and an independent foundation is planned.

Meta says agents using its Social Technologies MCP server can create apps and Business Portfolios and add use cases such as WhatsApp Business messaging.

Forever Security found flaws in the built-in browsers of Cursor, Copilot, Codex, VS Code and Claude Desktop. It says the Codex and VS Code flaws are unpatched.

Notion 3.8 lets outside AI agents use your workspace over MCP, connects Notion agents to 150+ tools and adds column permissions. What to set up first.

OpenAI describes a four-layer system that screens its agents for misaligned behavior and pages humans, plus its gaps: paging is off for internal Codex traffic.

MPO trains LLM agents with RL credit per turn, then per token. It topped GRPO, PPO and SAO on all nine reported metrics at 4B and 30B-A3B scale.
