ModelsNews
IBM released Granite Switch 4.1.1, a patch release of its modular model. The base model is untouched; only the decision adapter and the speech recognizer were updated, so integrations built on 4.1 keep working unchanged.
· 2 min read
ModelsNews
Baseten's research lab Base Labs released joey-1, a 2-billion-parameter decision model. Instead of text, it returns a single label with a calibrated probability, at 20 ms P50 latency and one cent per million input tokens. The weights are open under the MIT license.
· 2 min read
ModelsNews
AssemblyAI, known for its speech-to-text models, has released its first text-to-speech model. AssemblyAI-TTS-1 targets voice agents: $15 per million characters, unlimited concurrency and no contract.
· 2 min read
ModelsNews
Google DeepMind announced Gemini 4 Carbon, which it calls its strongest model yet. It arrives two weeks after Argon with a focus on cybersecurity and agentic tasks that run for days. For now, only a subset of cyber defenders gets access.
· 3 min read
ModelsNews
NVIDIA released Nemotron 3.5 Transcribe, an open-weight model that transcribes 40 languages, translates, labels speakers and answers questions about audio. It ships with the Granary v2 dataset and fine-tuning recipes.
· 3 min read
ModelsNews
Lausanne-based Giotto launched Alto and Solo on Tuesday as European sovereign models. Within hours its chatbot called itself Kimi, and the CEO admitted the model inherits some Kimi weights.
· 4 min read
ModelsNews
DiffuSpace released Clover-35B-A3B-Preview, an open diffusion language model under Apache 2.0. It generates text in parallel blocks and, according to the company, beats same-size Qwen models on agentic coding.
· 2 min read
ModelsNews
Ant Group's inclusionAI team released Ling-3.1-flash-1019 weights: a 560B MoE with 25B active parameters, 1M token context and an MIT license. It is clearly stronger than its predecessor on agentic tests.
· 2 min read
ModelsNews
Liquid AI released the LFM3 family: text, vision, audio, omni and decision models, all around 2B parameters and built on one shared base model. All are open-weight and fine-tune on a single GPU.
· 3 min read
ModelsNews
Google DeepMind has unveiled Genie 4, a world model that generates photorealistic, steerable worlds in real time with native audio and dialogue. For now, only trusted testers have access.
· 2 min read
ModelsNews
Baidu released the 3.34B-parameter Unlimited OCR 2 under the MIT license. Its new attention mechanism keeps decoding latency and memory nearly constant, no matter how long the document is.
· 2 min read
ModelsNews
vLLM Semantic Router released d3-ultra, a 27-billion-parameter multimodal decision model under Apache-2.0. It does not generate text, but returns a probability for every possible answer.
· 2 min read
ModelsNews
OpenAI has released GPT-6.1 Terra, a fast, low-cost model for computer use, visual reasoning and writing. Its API price is half that of GPT-5.6 Terra, and it is available today in ChatGPT Work and Codex.
· 2 min read
ModelsNews
The Chinese Nanbeige team released Nanbeige4.3-33B under Apache 2.0. Its own evaluations show it leading most agentic coding and office benchmarks among open models around 30B.
· 2 min read
ModelsNews
Tencent Hy released Hy-Nomas, an interactive video world model, under Apache-2.0. You steer it segment by segment with text instructions and camera moves, and it runs on a single RTX 4090.
· 2 min read
ModelsNews
NVIDIA released Nemotron 3.5 Bolt E4B, an open small model built on Gemma 4 E4B for local voice assistants, game characters and robots. Quantized, it runs in about 4 GB of VRAM.
· 2 min read
ModelsNews
Moonshot AI released Kimi K3.1 Pro, a 2.8-trillion-parameter MoE model with a 1-million-token context window. It ranks 4th on the Artificial Analysis index, and the weights are now on Hugging Face.
· 3 min read
ModelsNews
Block's research lab unveiled Dessa, a full-duplex speech-to-speech model trained without any text, only on recordings of human conversation. It is a research preview with a waitlist.
· 2 min read
ModelsNews
Shanghai AI Laboratory released the weights of its Atria Dawn agentic model on Hugging Face under the MIT license. Built on the 744B-parameter GLM-5.2 MoE, it handles 512K context but accepts text input only.
· 2 min read
ModelsNews
Ideogram released its first vector model on October 22. Vector 1.0 turns a text prompt or an existing image into scalable, editable SVG artwork, in the app and via the API.
· 2 min read
