AI Briefing#open-weights
The latest AI and developer news about #open-weights, with the original source and a short summary.
The latest AI and developer news about #open-weights, with the original source and a short summary.
Alibaba has announced the upcoming release of the 2.4T-parameter Qwen3.8 model along with an open-weight release.
Moonshot AI has unveiled Kimi K3, a large-scale open-weight model with 2.8 trillion parameters.
Open-source AI has surpassed closed models in token throughput and matched them in coding ability, but production deployment still lags due to operational issues.
Thinking Machines has unveiled Inkling, a 975B-parameter multimodal open-weight model.
As the share of production-environment token processing shifts to open-weight models, open source AI is establishing itself as a substantial mainstream force.
LingBot World has been released, a model that solves the drift problem occurring during long-term generation through MoBA attention masks and distillation techniques.
Gemma 4 is a next-generation open-weight multimodal model with diverse architectures and innovative features.
Poolside has released Laguna XS 2.1, an MoE model optimized for agentic coding.
DeepReinforce released Ornith-1.0, a coding-specialized model built on Gemma 4 and Qwen 3.5.
An analysis shows that the pattern of the performance gap between Open Weights models and Closed Source models differs depending on the benchmark metric used.
Krea has released Krea 2, an open-weight, 12B-parameter image generation model optimized for creative exploration and control.
GLM-5.2, released by Z.ai, is an open-weight model that nonetheless delivers agentic performance competitive with top-tier models.
Mistral plans to launch a new model family this summer, maintaining its open-weight principles and running an early access program for research and industry partners.
Cohere has unveiled North Mini Code, a 30B-scale model optimized for code generation and agentic tasks.
The Apodex-1.0 Smol model series, designed to optimize agent verification and tool execution, has been released as open weights.
More than 25 new open-weight models have been released across various fields including LLMs, images, audio, and video.
NVIDIA has unveiled Nemotron 3.5 ASR, a roughly 600-million-parameter model that supports 40 languages in real time.
Ideogram has unveiled **Ideogram 4**, its first open-weight text-to-image generation model.
China's MiniMax has released M3, a multimodal model that captures both performance and cost-efficiency through its innovative MSA architecture.
NVIDIA has unveiled Nemotron 3 Ultra, the most powerful US open-weight model at 550B parameters.
MiniMax has unveiled **MiniMax M3**, an open-weight model that combines coding, 1M context, and native multimodal capabilities.
MiniMax plans to release M3, a multimodal model using Deepseek-style attention, as an open weight model.
NuExtract3, a VLM based on Qwen3.5-4B specialized for document data extraction and Markdown conversion, has been released.
Numind has released NuExtract3, a 4B-scale open-weight VLM specialized for Markdown conversion and structured data extraction from documents.