Bland’s Tofu-1 is a full-duplex speech-to-speech model; Bland says over 92% of screening listeners couldn’t tell it apart.

Cmpnd’s dspy-wsp brings weight-space programming to DSPy, letting GPT-6-Astra edit weights directly and lift d1-omni-600M ...

Stanford’s Memorilla memory system feeds frozen Qwen3-8B 16 memory tokens, beating full context on FactKG: 84.33% vs 80.43%.

Vercel UI SDK is an Apache-2.0 library of 50+ UI behaviors that coding agents compose into accessible components.

CopilotKit released OpenBot 1.0, an MIT-licensed, self-hostable AI coworker workspace that runs policy checks on every agent ...

The AI video leaderboard looks very different from 12 months ago. Native audio is now table stakes. Clips run 15–30 seconds. And an open-weights model sits in the top 3. This ranking (as of Nov 13, ...

Google’s Gemini Decisions API scores up to 1,000 multimodal questions per call with Gemini Decision 1 at $0.10/M.

Meta FAIR that updates centroids with exponential moving averages. How does e-means work? Every k-means variant keeps a sum S ...

Ricursive’s Ripple-65B Preview is a 65B dense open-weight model scoring 80.5 on Terminal-Bench 2.1, released under Apache 2.0 ...

Google has released Gemini 4 Hydrogen, a fast model for coding and agentic work. Gemini 4 Hydrogen is a proprietary model from Google DeepMind built for everyday agent loops. It sits below the ...

Liquid AI’s LiquidCodec audio codec: a 74M-parameter streaming speech tokenizer encoding 361x real time on one CPU thread.

Nous Research’s Arete-27B fine-tunes Qwen3.8-27B with 52.8M tokens, cutting propaganda answers from 22% to 0.9% on 803 ...