The NIL-3B project has published a GGUF build, nil-3b-q4_k_m.gguf, for running the model locally with a GGUF-compatible inference runtime. Users install dependencies and run `python -m src.chat`, which expects the GGUF file in the project root. The repo notes its LICENSE covers source code only and does not confirm license or redistribution rights for model weights or training data.
inclusionAI released Duo-1.0-distill-quant, FP8-quantized four-step distilled Realization weights for the Duo 1.0 full-duplex multimodal world model. The weights overlay onto the inclusionAI/Duo-1.0-causal-base bundle, which supplies the shared Intent, codecs, speech models and Realization backbone, and require a runtime supporting the packaged ModelOpt FP8 format. The weights were initialized from MiniMax H3 and then trained by the Duo team.
inclusionAI has released Duo-1.0-distill, the four-step distilled Realization weights for Duo 1.0, a full-duplex multimodal world model for real-time audio-visual interaction. The weights ship as a PyTorch Distributed Checkpoint overlay on the inclusionAI/Duo-1.0-causal-base bundle, using an Euler sampler with 4 steps and flow shift 5.0. The weights were initialized from MiniMax H3 and then trained by the Duo team.
Sing VL S1, a 4B-parameter multimodal decision model built on Qwen3.5, now has a GGUF release for llama.cpp CPU inference. It ships in BF16 (8.42 GB), Q8_0 (4.48 GB) and Q4_K_M (2.71 GB) decoders plus a shared F16 vision projector, with a default 4096-token context. GGUF validation covers 9 examples and 11 questions; full public scores for the GGUF variants are still pending.
Edge0 open-sourced Edge0-145B and Edge0-120B along with its unified inference framework and deployment tools, claiming Edge0-145B is the first open-source 145B-parameter on-device model. The company reports 29GB peak memory, 1,100+ tokens/s prefill, 80+ tokens/s decode and roughly 120W typical power on a compatible laptop, with 128K–256K context. It cites internal scores of 64.6 on DeepSWE 1.1, 86.52 on Terminal-Bench 2.1, 90.4 on GPQA-Diamond and 80.6 on IFBench, and says weights are on Hugging Face with day-one support for macOS, Windows and Linux.
Why it matters: Edge0 open-sources a 145B on-device model that reports 29GB peak memory and 80+ tokens/s decode, offering a concrete data point on how far local inference can scale.
Liquid AI released LiquidCodec, a roughly 74M-parameter streaming speech tokenizer that powers its LFM3-Audio-2B and LFM3-Omni-2B models, with weights published on Hugging Face. The codec compresses 24 kHz audio into 12 token streams at 12.5 Hz frames and runs on a single CPU thread, where Liquid AI reports about 361x real-time encoding and 87x decoding. Liquid AI says it beat X-Codec 2, NanoCodec and Mimi on ViSQOL and mel distance across all 8 test sets, though NanoCodec still led PESQ (WB) on 7 of 8 sets.
IBM released two experimental preview models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, derived from Granite 4.1 base checkpoints and trained to consume and generate LLMON, an LLM-native markup language that marks instructions, data and requirements with reserved special-token delimiters. On the OPI benchmark the models avoid 47% more indirect prompt injection attacks than the best frontier model tested and 95% more than the best open-source model, while performing on par with Granite 4.1 Instruct on standard benchmarks. The release includes a GitHub toolkit with lossless JSON-to-LLMON converters, and the authors note the preview is for exploration, not production.
Why it matters: IBM trained a preview model on Granite 4.1 that can read and write LLMON tags, using explicit instruction-data separation to reduce the indirect prompt injection success rate on the OPI benchmark to near zero while maintaining general capabilities.
MBZUAI's Institute of Foundation Models released PAN 2, a 29B-parameter open world model that takes a scene plus a text action and simulates the outcome as video, with weights, training code and inference code under Apache 2.0. It pairs an LLM planner that writes structured JSON world states with a 29B MMDiT diffusion transformer conditioned by a frozen Qwen3.5-9B encoder, and IFM reports 34.0% full-plan completion on 100 multi-step tasks, ahead of MiniMax-H3 at 32.50%, while trailing MiniMax-H3 on single actions and human preference (Elo 1595 vs 1620). IFM's ablation shows removing the LLM planner drops full-plan completion to 0.0% and step completion from 75.1% to 8.7%.
Tether AI Research released BotPsy, a 0.8B-parameter agentic model built on Qwen3.5-0.8B for on-device personal-assistant tool use. It roughly doubles its base model's agentic scores, with BFCL v4 rising from 23.4 to 44.0 and τ-bench from 23.2 to 51.5, and beats every model up to 1.2B parameters tested across six tool-use benchmarks. It also cuts unsafe responses to 21% from 36% and uses about 3× fewer reasoning tokens on GPQA Diamond.
inclusionAI released LemonDecider 4B v3, a decision model fine-tuned from Qwen3.5-4B that returns choice, yes/no or ordinal probabilities via a single joint forward pass instead of generating answer tokens. It was trained on about 1.20M decision/QA rows and scored 61.67 on the public Decision Index 0.4 suite, with median request latency of 49.9 ms.
inclusionAI has released GGUF-format weights for ZoomGain-9B, converted with llama.cpp, covering the language model and the vision encoder/projector (mmproj) in FP16, Q8_0 and Q4_K_M. Image input requires pairing the language model with the matching mmproj file, and the repo documents CLI inference via llama-mtmd-cli and a web chat setup via llama-server. A llama.cpp build with Qwen3.5 multimodal support is required, and quantization may affect accuracy.
Latent Space tested NaiveAI's open-weights Naive-N0.5-Flash, a coding model derived from Xiaomi's MiMo-V2.5 and paired with the NaiveRT inference engine, against DeepSeek V4.1 Flash 1017 and GLM 5.3 Flash. In the author's runs it peaked near 2,000 tokens per second and sustained about 1,000 during thinking-heavy tasks, with a median response time of 3.1 seconds versus 5.3 and 10.1 seconds for the two comparison models. On hard HLE questions it averaged about 108,500 completion tokens with thinking on, roughly nine times GLM 5.3 Flash and 11 times DeepSeek V4.1 Flash 1017, and the author notes the sample was 10 questions per difficulty level with no accuracy grading.
MiniMax released H3-Image, an open-weight image generation model built on H3 that supports text-to-image, reference-based generation and image editing, and can produce a 4096 × 4096 image in under 10 seconds. The company says it handles up to 16 reference images, claims state-of-the-art prompt adherence among open-weight models, and reports speeds up to 5× mainstream alternatives. H3-Image is available via the MiniMax API Platform, MiniMax Code and Hailuo AI, priced at 50–70% of comparable offerings, with 50% off 4K-and-above requests during launch week.
Why it matters: MiniMax releases open-weight H3-Image, claiming sub-10-second 4K generation and pricing at 50–70% of comparable mainstream offerings, giving developers a new open option for high-resolution image work.
Amazon introduced Mitra-v2, a second-generation tabular foundation model for classification and regression, released with fine-tuned models and inference and fine-tuning code under the Apache-2.0 license on Hugging Face. It is pretrained on synthetic tables with 10 times as many rows and roughly three times as many columns as Mitra-v1, using a new hybrid SCM prior and hybrid optimizer while keeping the Transformer architecture unchanged. On TabArena, a public benchmark of 51 real-world datasets, Mitra-v2 leads the single-model field, beats Mitra-v1 on 32 of the 33 datasets both run natively, and finishes all 51 within the one-hour budget; fine-tuned Mitra-v2 also ranks first on ScoringBench.
Why it matters: Mitra-v2 shows that widening synthetic pretraining data, rather than deepening the network, lifts a tabular foundation model to the top of TabArena's single-model field.
Zyphra released ZONOS3-Mini, a sparse MoE text-to-speech model with 100M active and 800M total parameters under Apache 2.0, trained on 112 NVIDIA H100s in 2.5 days. It reaches under 570 ms time-to-first-audio in the browser on a Pixel 9 and under 150 ms on an M3 MacBook Air, with runtimes for WebGPU, SGLang, vLLM and GGML. The model adds control over identity, accent, quality, mood and delivery, including word-level steering, and Zyphra reports it ranks first among open models on its new ZTTS2-Eval benchmark.
Why it matters: Zyphra's compact MoE TTS model pairs zero-shot voice cloning with word-level delivery control while running in real time on phones and laptops.
Preferred Networks released PLaMo 3 Translate, a translation-specialized LLM used in its PLaMo Translation service, as an open-weight model on Hugging Face on November 5. 8B, 2B, 8B and 31B) and supports 53 languages including Japanese and English, running on smartphones, laptops and internal servers without a cloud connection.
Why it matters: PFN releases a translation-specialized LLM in four sizes from 0.8B to 31B, letting companies and individuals run 53-language translation on-device without cloud access.
Nanbeige has released Nanbeige4.3-33B, an open-weights dense reasoning model under Apache 2.0 that scores 20 on the Artificial Analysis Intelligence Index v5, matching Qwen3.8 27B (xhigh) at 20. The 33B model has a 524K-token context window, text-only input and output, and no confirmed public APIs, and it scores 30% on Terminal-Bench 4.0 versus 6% for Qwen3.8 27B while trailing on HLE-Diamond (7% vs. 15%). On AA-Omniscience v1.1 it records an 18% hallucination rate versus 29% for Qwen3.8 27B, with a 71% decline rate, and it uses 158K output tokens per Intelligence Index task versus 110K for Qwen3.8 27B.
Mistral AI released the open weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter natively multimodal model that entered public preview earlier this month and was trained from scratch in Europe. The company says three additional weeks of reinforcement learning lifted it to the top of open-weight coding, with 41% on Terminal-Bench 4.0 and 48% on the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task than GLM-5.3. It adds Low, Medium and High reasoning modes, scores 51% on the Artificial Analysis Cyber Index, and is priced at $0.70 input, $0.07 cached input and $2.10 output per million tokens under a modified MIT license.
Why it matters: Mistral opens the weights of its 1-trillion-parameter flagship, pairing coding and defensive-security benchmark gains with three reasoning modes and per-token pricing.
inclusionAI released Ling-3.0-flash-Sante-fp8, an FP8-quantized version of its healthcare-specialized Ling-3.0-flash-Sante model, keeping 124B total and 5.1B active parameters per token. It matches the BF16 checkpoint's 73.9 average across healthcare benchmarks while cutting weight memory for FP8-capable hardware. The research-only model supports 256K-token contexts and defaults to thinking mode with tool calling.
Ai2 released OlmoEarth v1.3, a satellite embedding model that turns a year of Sentinel-1, Sentinel-2 and Landsat time series into one 128-dimensional vector per 10-meter pixel, with the first 64 dimensions usable as a smaller embedding. Ai2 precomputed annual embeddings for the whole planet and published them on Source Cooperative, and its 128-dimensional embeddings scored highest on 20 of 27 evaluations against AlphaEarth Foundations and Tessera v2, including a new tropical crop benchmark called PLANTEUR. The release also includes the OlmoEarth Embeddings Explorer, a no-sign-in browser tool for comparing years, finding similar places and training small classifiers on the spot, with 2025, 2024 and 2023 global and 2017 to 2022 covering select countries; weights, pretraining data and code are open.
inclusionAI released Ling-3.0-flash-Sante, a healthcare-specialized model built on Ling-3.0-flash with 124B total parameters and 5.1B active parameters per token, with weights on Hugging Face and ModelScope. It uses a hybrid-linear MoE architecture combining Kimi Delta Attention, gated MLA and sparse MoE layers, trains to 256K tokens, and enables thinking mode by default with tool calling through compatible serving frameworks. The company labels it a research model, not a medical device, and says outputs must not replace professional medical advice, diagnosis or treatment.
inclusionAI released Ming-3.0-tts, an open-weight multilingual speech synthesis family with two models: Ming-3.0-tts-0.5b, a 0.5B dense model, and Ming-3.0-tts-tiny, a MoE model built on Ling V3 with 7.9B total and 1.3B active parameters. Both ship with inference code, a task cookbook and benchmark tables, and support voice cloning, text-based voice design, free-form natural-language delivery control, 14 inline paralinguistic tags, emotion and speed/pitch/volume control, plus Chinese, English, German, Spanish, French, Korean, Japanese, Sichuanese and Cantonese. On the October 2026 Artificial Analysis text-to-speech leaderboards, Ming-3.0-tts ranks 1st among open-weights models on Controlled Voice, 1st overall on Pronunciation Robustness at 93.2%, and 3rd among open-weights models on Provider Voice.
inclusionAI released Ming-3.0-asr-tiny, a speech recognition model that pairs the Ling-tiny-v3 language model with an audio encoder for Chinese and English transcription, dialect recognition and context ASR. Dialect evaluation covers 15 test sets including Hunan, Minnan, Cantonese, Chuanyu and Shanghai, and hotwords or preceding dialogue can be supplied in the prompt to resolve names and technical terms. The repository ships Transformers and SGLang serving guides, and code and weights are released under the MIT License.
Mistral released Timestral, its first family of time-series foundation models, in four sizes from 5M to 209M parameters. The largest model ranks first on all three main public benchmarks, GIFT-Eval, TIME and fev-bench, among pre-trained zero-shot models from a single checkpoint without ensembling, using two-thirds of the parameters of the previous state of the art. Timestral produces probabilistic multivariate forecasts with 21 quantiles, supports past-only and known-future covariates, runs on CPU without a GPU, and is available through Mistral Forge for fine-tuning, with a GitHub repository and technical report published under a modified MIT licence.
Why it matters: Mistral's first time-series foundation model family claims top zero-shot rankings on GIFT-Eval, TIME and fev-bench with a 209M-parameter checkpoint, offering a size-accuracy tradeoff reference for forecasting work.
inclusionAI released Ming-3.0-audio-flash, a 126B-parameter Mixture-of-Experts audio model with 6.5B activated parameters, alongside the compact Ming-3.0-audio-tiny at 9.3B parameters with 2.7B activated. The flash model covers Chinese and English speech recognition across dialects and noisy multi-speaker settings, long-audio understanding, non-speech sound and music perception, audio reasoning, and end-to-end spoken dialogue with optional synthesized speech output. It ships with vLLM-Omni, SGLang, and Transformers inference paths under the MIT License, with default Flash profiles using tensor parallelism across four 80 GB H800 GPUs at 128K context.
Why it matters: The release details a 126B-parameter MoE audio model with 6.5B activated parameters, plus a 9.3B tiny variant, giving a concrete reference for open audio-language model sizing and deployment.