Alibaba’s Qwen4-Max is a 2.4T-parameter MoE model with 71B active, native vision, 1M context, and $1.5/M input pricing ...
Hark releases Handoff 2, a computer-use agent that runs 2x faster than v1 and completes 82% of 478 internal web tasks.
Meta has moved Catching Just-in-Time Tests (JiTTests) from research prototype into its production infrastructure. Meta JiTTests is an LLM-based system that writes executable tests designed to fail on ...
Bland’s Tofu-1 is a full-duplex speech-to-speech model; Bland says over 92% of screening listeners couldn’t tell it apart.
Vercel UI SDK is an Apache-2.0 library of 50+ UI behaviors that coding agents compose into accessible components.
Cmpnd’s dspy-wsp brings weight-space programming to DSPy, letting GPT-6-Astra edit weights directly and lift d1-omni-600M ...
Google’s Gemini Decisions API scores up to 1,000 multimodal questions per call with Gemini Decision 1 at $0.10/M.
Stanford’s Memorilla memory system feeds frozen Qwen3-8B 16 memory tokens, beating full context on FactKG: 84.33% vs 80.43%.
Liquid AI’s LiquidCodec audio codec: a 74M-parameter streaming speech tokenizer encoding 361x real time on one CPU thread.
CopilotKit released OpenBot 1.0, an MIT-licensed, self-hostable AI coworker workspace that runs policy checks on every agent ...
AlphaGenome 2 from Google DeepMind predicts variant effects per cell type, recovering 30% of causal sc-eQTLs versus 16%.
Meta FAIR that updates centroids with exponential moving averages. How does e-means work? Every k-means variant keeps a sum S ...