Perplexity releases Decider v1.3 and Decider Flash v1.3, general-purpose decision/classification models returning probability distributions over answers for binary, multi-class, and scoring tasks, with improved robustness, bidirectional attention, and a new 'locate' feature for NER and extractive QA.
World Labs launches Atlas, a next-generation world model that turns images into explorable 3D worlds, with tools like Viewfinder and Chisel plus developer APIs for converting splats to meshes for games, film, architecture, and robotics workflows.
Google introduces Gemini 4 Hydrogen, a fast and efficient Gemini 4 model built for everyday coding and agentic workflows, priced flat at $1.50/$7.50 per million tokens across its 1M-token context window and rolling out across Google AI Studio, the Gemini API, the Gemini app, Search, and Google Antigravity.
IBM releases LLMON, an LLM-native markup language, along with two experimental Granite 4.1 3B and 8B preview models trained to consume and generate LLMON-marked text, improving structure handling and nearly eliminating indirect prompt injection attacks.
In its podcast, Google DeepMind introduces Alpha Genome 2, advancing AI's ability to read genomes down to the level of individual cell types — enabling researchers to score and rank the impact of single DNA mutations and to pinpoint truly noteworthy variants across the 98% of the non-coding genome.
Haru-4B is a low-latency, multilingual full-duplex continuous spoken dialogue foundation model that listens and speaks simultaneously at 7.5 Hz, using a Qwen3.5-4B backbone with a MeanFlow audio head for one-step continuous audio generation. It is released as a research base model with PyTorch, MLX, and vLLM-Omni implementations.
Edge0 released Edge0-145B-NVFP4, a 145B latent sparse MoE agent model natively trained at 4-bit weights that runs on a single RTX 5090 24GB consumer GPU with 98.3 tok/s decode, 16K prefill of 3165.7 tok/s, and 86.5 on Terminal-Bench 2.1, using VRAM+host RAM residency for expert streaming.
Abiray/MiniMax-H3-Image-GGUF 发布 MiniMax-H3-Image(33.1B 单流 DiT)的 GGUF 量化权重,敏感层保留 FP32 并原生支持 ComfyUI-GGUF 加载,覆盖 Q2_K 至 Q5_K_M 多档位以适配 8–24GB VRAM。
Thinking Machines Lab released Inkling 2 Small, an open-weights general-purpose multimodal model (text+image input, text output) aimed at developers building agentic, coding, chatbot and RAG applications, with NVFP4/MXFP8 checkpoints and broad inference and post-training toolchain support.
ChatR Labs released a quantized, uncensored Omni adapter merge of Qwen-Image-2.1 for text-to-image generation and image editing, shipping transformer weights in multiple quant formats (GGUF, FP8, INT8, NF4, BF16) with recommended inference settings for ComfyUI and Diffusers.
Particle-1, a coding-agent-focused model emphasizing token efficiency (~45% fewer output tokens than DeepSeek V4.1 Flash with comparable accuracy), is now available in OpenClaw v2026.10.10, with zero data retention for standard inference.
Skyfall AI unveiled SKY-001, an engineering world model that reconstructs a Jaguar car with thousands of 3D parts, claiming CAD accuracy on par with GPT-6 Astra while running 51× faster and costing 150× less.
OpenCode has launched a mysterious stealth model, Pulse Alpha, offering 1M context, 131K max output, multimodal input, and three reasoning tiers — free for a limited time. Speculation points to GLM 5.5 or the Qwen series under the hood.
balbal-v2 releases two CPU-efficient Turkish speech recognition models (313M and 45M params, CTC + KenLM) trained for under $250, ranking 2nd on the Mu-Bench TR benchmark behind Google Chirp 3 while running at up to 236x real time on modest hardware.
Deep Principle released Hydrogen-1.0, an open-weight model post-trained from GLM-5.3 for chemistry, materials science, and life sciences, showcased via a short film of it embedded in a daily research workflow alongside their Mira AI Scientist assistant.
NVIDIA released NemotronLabs RE-USE, a 9.6M-parameter multilingual universal speech enhancement model that restores speech quality across diverse degradations and sampling rates, outperforming larger open-source USE models while being ready for commercial use.
Fable 5.5 reportedly generated a complete animation in a single pass using only code. Chetaslua said that standard Opus 5.5 and Fable 5.1 produced noticeably worse results with the same prompt and provided a method for verification, but the report has not yet been officially confirmed.
Maxime Labonne celebrates reaching 1,000 likes in five days for his post about the LFM3-1.7B small language model.
Recommends the locally deployable face swap LoRA "BFS-Best-Face-Swap," supporting face, head, and body replacement while preserving the original hair, skin texture, and lighting. Provides ComfyUI workflows, Draw Things presets, and Python Diffusers deployment options.
A user reports that Qwen 3.8 27b, a local model, autonomously discovered it could access Blender via MCP and wrote Python code to speed up its math on a physics problem — behavior the user says they've never seen from a local model before.