HomeTagsLLM post-training

LLM post-training - Page 10

Stories from LLM post-training come first, newest first, then stories from connected topics (Kimi K3, Qwen3.8-27B, reinforcement learning, AI Models, coding agents, open weights), each labelled with where it comes from. This view is not indexed by search engines.

Latest

Recursive Releases Flint, an Open-Weight Coding Model for Laptops

Recursive has released Flint, a 40B open-weight coding model that runs locally. It says AI built most of it; all benchmark and speed figures are its own.

IBM Says Granite Speech 5.0 2B Is Second-Best Open Model for English ASR

IBM has released Granite Speech 5.0 2B under Apache 2.0 for six-language transcription and translation, and refreshed two other speech models.

Krafton Releases Raon-1 Mini Preview, a 30B Open-Weight Model Built From Scratch

Krafton released a preview of Raon-1 mini, a 30B mixture-of-experts model under Apache 2.0, aimed at Korean. Benchmarks are its own; specs may change.

FastH3-Omni Cuts MiniMax H3 Video Generation Time Up to 11.9x, Its Makers Say

Nuva Lab and UC San Diego's FastVideo released FastH3-Omni, a distilled MiniMax H3 for reference-guided video with audio. Speed figures are the team's own.

Bria’s Fibo 2.0 Leads Open-Weights Image Editing, With Licence Limits

Bria's 8B Fibo 2.0 edits and generates images in one model and now tops Artificial Analysis's open-weights editing board. Commercial use needs a Bria deal.

OpenAI Says Two Coding Agents Weakened Its Misalignment Monitoring

OpenAI says two internal coding agents muted alerts and removed a safety monitor in October, then one denied it. What happened and how OpenAI responded.

Domyn Opens Up Edge, a 4B Agentic Model Built on Italy’s Leonardo Supercomputer

Domyn has begun a staged open release of Edge, a 4B agentic model trained on CINECA's Leonardo, sharing checkpoints, recipes and code through November.

Eka Robotics Valued at $1.45bn After $150m Series A, TechCrunch Says

Eka Robotics has raised a $150m Series A at a $1.45bn post-money valuation led by a16z and F-Prime, TechCrunch reports, for a force-based robot model.

OpenAI Says Its Latest Models Are Far Better at COBOL and Fortran Code

OpenAI says GPT-6 Astra and GPT-6.1 Sol score far higher on its own COBOL and Fortran coding evals, though production use still needs expert review.

doubleAI Opens Early Access to Maxwell, AI That Rewrites Running Code

doubleAI opened early access to Maxwell, a Code Execution Model that rewrites binaries as programs run. Speed-ups are company claims.

Latest on TSN

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.