#open weights
Open Weights: 15 AI articles covering open weights news, analysis, and research
Articles
Baidu Releases ERNIE 5.5 Flash Base: A 280B MoE Model With 14BNEWโญ9
Baidu releases ERNIE 5.5 Flash Base, a 280B MoE model with 14B active parameters, 128K context, and top scores on 24 of 30 base-model benchmarks.
Baseten Releases joey-1: A 2B Decision Model Priced at $0.01 perโญ9
Baseten's joey-1 is a 2B decision model that returns labels with calibrated probabilities for routing and classification, priced at $0.01 per 1M input tokens.
Welcome Le Chonk: Mistral Large 4 on Hugging Faceโญ8
Mistral Large 4, nicknamed Le Chonk, is now on Hugging Face: a 1T-parameter MoE multimodal model with 52B active parameters, built for image-text-to-text tasks.
Embedded AI Evaluators Won't Be an Easy Fix, Says Hugging Faceโญ9
Hugging Face's Irene Solaiman says AI model evaluations remain immature and hard to reproduce, and embedded evaluators need far more standardization before they...
Mistral AI Releases Timestral: A 209M-Parameter Forecastingโญ10
Mistral AI releases Timestral, a family of four zero-shot time series models up to 209M parameters that runs on CPUs and beats Google's TimesFM-3 on key benchma...
Kyutai REKI: A 2.4B Vision-Language-Action Model Bringingโญ8
Kyutai REKI is a 2.4B vision-language-action model for robot manipulation, pretrained on public data with efficient memory and strong out-of-distribution perfor...
StepFun Releases Step 5 Preview as Open Weights: 600B MoE, 27Bโญ9
StepFun releases Step 5 Preview open weights: a 600B MoE model with 27B active parameters, 1M-token context, and strong long-context reasoning for agentic workl...
How AI Decision Models Could Transform Content Moderationโญ9
Musubi's PolicyLM-1.7B open-weight AI decision model applies plain-English content policies in under 50ms without retraining, transforming real-time content mod...
The Story of Qwen: Alibaba's AI Models From 7B to 2.4Tโญ7
From a 7B chatbot in 2023 to open-weight models with 2.4 trillion parameters, here's the story of how Alibaba's Qwen family evolved release by release.
StepFun Launches Step 5 Preview: A 600B-Total, 27B-Active MoEโญ8
StepFun's Step 5 Preview is a 600B-parameter MoE model with 27B active parameters and 1M context, built for long-horizon agentic work like coding and finance.