Hugging Face Explores Multi-Harness RL for More Portable Agents
Hugging Face researchers tested multi-harness reinforcement learning, training an open-weight model across multiple agent interfaces like Claude Code, Codex, an…
Global AI Digest
-
-
Security-One: Open-Weight 27B Model for Security Decisions Released
An open-weight 27B model called Security-One has been released for making security decisions, returning calibrated probabilities for user-defined answers.
-
Perplexity's Decider v1 27B Model Now Available on Requesty
Perplexity's new Decider v1 27B model is now available on Requesty, designed for fast decision-making tasks like routing, classification, and control logic.
-
Turning Real Coding Harnesses into RL Environments via a Proxy
A new approach turns coding harnesses like Claude Code, Codex, Hermes, Pi, and opencode into RL environments using a capture proxy that speaks the four API form…
-
RL Environments Get a Unified Home on the Hugging Face Hub
RL environments from OpenEnv, Verifiers, Harbor, NeMo Gym and more can now be discovered on the Hugging Face Hub, addressing the fragmentation where each RL fra…
-
RL Environments Get a Home on the Hugging Face Hub
Hugging Face is bringing reinforcement learning environments to the Hub, treating them like datasets so any framework can load them without a separate registry.
-
Ling 3.1 Flash Now Available Free on Requesty
AntLingAGI's Ling 3.1 Flash model, hosted by Novita Labs, is now available for free through the Requesty AI gateway.
-
DeepSeek Harness Plugin Authors Get New Upgrade Skill to Adapt to DSH Version Changes
A new skill called dsh-plugin-upgrade-skill is recommended to help DSH plugin authors update their plugins for new DSH versions, alongside a broader call for co…
-
Command Code GOAT Plan Includes $60 Credits for DeepSeek, GLM, and Kimi Models
Command Code lists four models included with $60 usage credits under its $10/month GOAT plan, covering DeepSeek V4.1 Flash variants, GLM 5.3 Flash, and Kimi K3.
-
Codex Team Pledges Daily Ship or Reset Over Next 28 Days
The Codex team says over the next 28 days it will ship either a clear improvement for most Codex/ChatGPT users or a full reset each day, focusing on simplificat…
