Guides - Page 17

Hugging Face's FineEnvs team has released an open framework that turns unmodified coding-agent harnesses such as Claude Code, Codex and OpenCode into reinforcement-learning environments. In its own tests, one small model scored 62% in one harness and 33% in another. What a harness is, how the training works, and what the results do and do not show.

An NVIDIA tutorial shows OpenAI's Codex coding agent turning a Jetson Orin Nano into a self-starting kiosk that captions a live webcam feed with Google's small Gemma 4 E2B model, running on the device. What the demo shows, what "No Mac required" means, and why local inference matters.

A new benchmark, TasteVal, says the best AI model now reaches expert-level results on eight AI research tasks with 2.3 times less compute, though its authors list reasons that may overstate progress. Preference Model has open-sourced Karotte, a framework for training environments that are harder to game, and DatologyAI has released Zephon, a data loader that keeps experiments repeatable when the GPU count changes. What each claims, and who measured it.

Scale AI and Elorian's Humanity's Sixth Sense benchmark finds people score 93.1% on everyday visual questions while the best model, GPT-6 Astra, scores 53.6%, mostly because models misread the picture. A separate preprint, not yet peer reviewed, says base models can reason almost as well as RL-trained ones when their answer starts with the right word.

Cognition has given its Devin coding agent a memory: short notes kept in a Git repository and tidied each night by a "dreaming" session, with the format open-sourced. Separately, Hark has released Hark Pro, a free assistant built on a model trained to use computers. What each launch is, and which claims are the companies' own.

A critical Langflow flaw let MCP server settings run any command. Researchers say GitHub's Copilot CLI followed encrypted instructions that it refused in plain text; GitHub says that is not a vulnerability. Meanwhile Atlassian has split its 220+ agent tools into read, write and destructive tiers, and OpenAI has made Codex Auto-review free. What each one confirms.

A hijacked release of the tensorlake npm package stole tokens and AI-tool config files, and booby-trapped the stolen GitHub token so that revoking it wipes the victim's home folder. The same week, GitHub made local sandboxing for Copilot generally available, Bitdefender released a free agent guard for macOS, and Citadel AI opened early access to agent monitoring. LangChain and past.dev shipped agent tooling with access controls built in. What each one confirms.

On 6 October OpenAI launched a Decisions API that returns yes/no probabilities, picks and rubric scores from GPT-6 Luna, and cut its API usage tiers from five to three. Perplexity released pplx-decider-v1.1-27b, an Apache 2.0 open-weight decision model. What decision models are, what each costs, and which scores are the vendors' own.

Microsoft made its Execution Containers generally available on Windows 11 as it rebuilds Windows for local and cloud "hybrid" AI, and opened Surface Laptop Ultra pre-orders from $2,599. AWS open-sourced Strands Box, a sandbox that enforces rules such as "no git push" whatever the agent decides. What an enforced sandbox does, in plain words.

Nous Research has raised $90 million to build Hermes for Businesses, a company version of its MIT-licensed Hermes agent. The raise is confirmed; the valuation is reported, mostly at $1.5B, with one $1.2B figure in circulation. Nous's "2.5% of global token usage" is its own estimate.