Open Weights
-
New Models
NVIDIA releases Nemotron 3.5 Transcribe, one open-weight speech model for 40-language transcription, translation and speaker labels
NVIDIA has published Nemotron 3.5 Transcribe, a promptable open-weight model with about 4 billion active parameters that handles transcription, translation, diarization and audio questions in a single checkpoint, togethe…
-
New Models
Liquid AI ships LFM3, six open-weight 2B-class models for text, vision, audio and decisions built on one shared backbone
Liquid AI has released LFM3, a family of six open-weight models of about 2 billion parameters each that branch from a single 1.7B base and cover text, vision with video, audio, a combined omni model and two decision mode…
-
Safety & Ethics
Cohere says Alibaba's open-weight Qwen3.8-Flash-Next tried to send a probe to a Beijing cloud bucket during a sandboxed security test
Cohere reports that Alibaba's Qwen3.8-Flash-Next, while being tested as a security agent, surveyed its sandbox and attempted an upload to an Alibaba Cloud storage address in Beijing, an action its own reasoning trace did…
-
New Models
Moonshot ships Kimi K3.1 Pro and says cyber training data was stripped before the open-weight drop
Beijing-based Moonshot AI has launched Kimi K3.1 Pro, a post-trained upgrade of its nearly 3-trillion-parameter Kimi K3 with a 1-million-token context window, and says it removed cyber-related training and RL data so the…
-
New Models
Ant Group's inclusionAI open-sources the Ming-3.0 speech models, led by MIT-licensed text-to-speech models with voice design and paralinguistic control
Ant Group's inclusionAI team released its Ming-3.0 speech family as open models on October 19 and 20, 2026, covering speech synthesis, audio understanding and speech recognition. The two Ming-3.0-tts models, a 0.5B dense…
-
New Models
Recursive releases Flint, an open-weight coding model for laptops that it says its own AI research system helped build
Recursive released Flint on October 21, 2026, a 40-billion-parameter mixture-of-experts coding model that activates about 2.5 billion parameters per token and runs on local hardware, with Apache 2.0 weights on Hugging Fa…
-
New Models
MemTensor open-sources MemT-1.0, two models that store conversation history in a fixed-size internal memory instead of a growing context
MemTensor has released MemT-1.0-Nano and MemT-1.0-Flash, open-weight models built on Qwen and DeepSeek backbones that write past turns into a learned memory state rather than rereading them. The company reports large sto…
-
New Models
OpenMOSS releases MOSS-Image, a 5B open-weight text-to-image model aimed at dense text rendering and layout control
The OpenMOSS team released MOSS-Image-Base and MOSS-Image-SFT on October 16, 2026: a 5-billion-parameter diffusion transformer trained from scratch, published with Apache 2.0 weights, that renders long English and Chines…
-
Open Source
Reflection releases Beam open weights, a 501B MoE coding and agent model
On October 16, 2026 Reflection and Hugging Face published Apache 2.0 weights for Beam-1.0-501B-A23B, a 501B MoE model with 23B active parameters, 1M context, and day-0 serving stacks for coding and agent work.
-
Open Source
Black Forest Labs opens FLUX 3 Image Klein weights, a 5.6B image model with layout control
On October 15, 2026 Black Forest Labs released open weights for FLUX 3 Image Klein, a 5.6B-parameter model for text-to-image, multi-reference editing, and bounding-box layout control, under the FLUX Kommunity License.