Home / New releases / Open-weight models

🔓 New open-weight and open-source AI releases

33 verified AI model and tool releases with downloadable weights or open-source code, newest first and grouped by month. Each entry links to the official announcement or model card, with a link to its own SIloadout release page. Last updated 2026-11-20.

Open weights do not always mean an OSI open-source license: check each summary and the model card for license terms before commercial use.

Jump to month: November 2026 (21) · October 2026 (12)

November 2026

All 83 releases from November 2026 →

2026-11-20

Hy4 open weights (Tencent)

Tencent published the open weights for Hy4, the natively multimodal flagship mixture-of-experts model it launched by API on Nov 4: 770B total parameters with 49B active per token (plus a 1.23B vision encoder and a 3.3B DFly2 speculative-decoding drafter), 78 layers, a 1M-token context window, text, image and video understanding, and computer use. The release includes Hy4 (instruct), Hy4-FP8 and the pre-trained Hy4 base model on Hugging Face, ModelScope and GitCode, with GGUF builds from AngelSlim, day-0 vLLM support and a System Card with safety evaluations. It uses the Tencent Hy License 1.0, which is derived from Apache 2.0 but is not Apache 2.0: model-as-a-service businesses with more than US$50 million in annual revenue or products with more than 100 million monthly active users need a separate written license, while purely internal use is exempt.

Source ↗
2026-11-19

Aion Instruct open weights and Edge Prompt API origin trial (Microsoft)

Microsoft made Aion Instruct (Aion-1.0-Instruct-2.3B), a 2.3B-parameter on-device small language model from its Windows Applied Sciences group, generally available as open weights on Hugging Face under the MIT license and as Microsoft Edge's built-in language model, replacing Phi-4-mini. It handles summarization, rewriting, chat and structured output in eight languages with a 32K-token context. Edge also opened an origin trial of the Prompt API so live sites can use the model on-device, plus a developer preview of an experimental Decisions API. Microsoft says Aion Instruct is also coming to Windows.

Source ↗
2026-11-19

Theorizer in Asta, open source (Ai2)

Ai2 launched Theorizer in Asta, its platform of AI agents for scientific research. Theorizer searches the literature, extracts findings from papers and combines them into testable theories, linking each claim to the supporting studies. Ai2 also open-sourced the code (asta-theorizer, Apache 2.0) so researchers can run it locally with their own model API keys; in Asta, Ai2 covers token costs up to a limit. Scientists have tested it for nearly a year, and Ai2 cautions that it can miss relevant studies and misjudge originality.

Source ↗
2026-11-19

Olala mini 7A1B open weights (OVHcloud)

OVHai LLM (OVHcloud group) released Olala-v1.0-mini-7A1B, the first model in its Olala family, with open weights under Apache 2.0 on Hugging Face. It is a small reasoning model for chat, instruction following and tool use (6.8B total parameters, about 1B active per token, hybrid Mamba-3 / attention mixture-of-experts, 32K context with a 128K version coming soon), trained on three European public supercomputers on 12.8T tokens, and it can run on a laptop or CPU. Base, SFT, summarise and translate variants were released alongside it, and it is hosted on OVHcloud AI Endpoints and Hugging Face Inference Providers.

Source ↗
2026-11-18

Hy4 flash open weights (Tencent)

Tencent released open weights for Hy4 flash, the low-cost everyday model it launched by API on Nov 16: Hy4 flash, Hy4 flash-FP8 and Hy4 flash-Base on Hugging Face and GitHub, with GGUF builds from AngelSlim and day-0 vLLM support. The model card lists 220B total parameters (including 64B Engram parameters) with 12B active per token, native text and image input and a 1M-token context window. It is released under the Tencent Hy License 1.0, which is derived from Apache 2.0 but adds a separate-license requirement for large model-as-a-service businesses.

Source ↗
2026-11-12

Inkling 2 Small open weights (Thinking Machines Lab)

The full weights of Inkling 2 Small are now on Hugging Face under Apache 2.0, after external safety testing. It is a 594B-parameter mixture-of-experts model (17B active) with image input, variable effort and a 1M-token context, launched Nov 5 on Tinker with closed weights. It is integrated with Unsloth, SGLang, vLLM and TokenSpeed, can be post-trained with Miles, SkyRL and Snowflake Cortex Training, and is available by API from NVIDIA, Fireworks, Together and others.

Source ↗
2026-11-12

Edge0-145B and Edge0-120B open weights (Edge0)

Open-weight on-device models for laptops, released with Edge0's unified inference framework and deployment tools. Edge0-145B is a sparse mixture-of-experts model with about 10B parameters active per step. Edge0 reports 29GB peak memory, 80+ tokens/s decode and a 128K–256K context on a compatible laptop, with day-one support for macOS, Windows and Linux (Apple M-series and NVIDIA RTX). The benchmarks are Edge0's own (e.g. 64.6 on DeepSWE 1.1). Weights are on Hugging Face (NVFP4 and q3k builds) under the Edge0 Community Model License 1.0, which is not a standard open-source license. An Edge0 App runs offline without an account.

Source ↗
2026-11-12

SOV-Transcribe-1 open weights (Salesforce)

Salesforce's first open-weight speech-augmented language model for non-streaming speech recognition, with 26B total and 4B active parameters. Weights are on Hugging Face under Apache 2.0. It targets best-in-class English transcription with multilingual support, entity understanding (phone numbers, emails), robustness to accents and noise, style steering and context biasing. Salesforce says it ranks #1 among open-weight models for non-streaming speech-to-text on Artificial Analysis (#7 overall) as of Nov 12, 2026.

Source ↗
2026-11-10

MiniMax H3-Image open weights (MiniMax)

Open-weight image model built on MiniMax H3 for text-to-image, reference-based generation (up to 16 reference images) and editing, generating native 4K (4096x4096) images in under 10 seconds. Weights are on Hugging Face (MiniMaxAI/MiniMax-H3-Image) under a community license dated Nov 10 that excludes the EU, UK, South Korea and the US; it is also available through the MiniMax API, MiniMax Code and Hailuo AI, with 50% off 4K-and-above requests during launch week.

Source ↗
2026-11-09

LongCat-2.5 open weights (Meituan)

One week after launching LongCat-2.5 (Nov 2), Meituan released its weights on Hugging Face under the MIT License, which allows commercial use, modification and redistribution. The natively multimodal mixture-of-experts model has 1.6T total parameters (about 48B active per token) and targets long-horizon agentic coding, visual coding and 3D tasks. A W4A4 quantized checkpoint is also published.

Source ↗
2026-11-09

ZONOS3-Mini open weights (Zyphra)

Zyphra released ZONOS3-Mini, a compact zero-shot voice-cloning text-to-speech model: a sparse mixture-of-experts with 100M active and 800M total parameters under Apache 2.0, with controls for identity, accent, mood and word-level delivery. It runs in real time on-device (under 150 ms to first audio on an M3 MacBook Air) with runtimes for browsers via WebGPU, SGLang/vLLM and GGML. Weights are on Hugging Face, and it is also served on Zyphra Inference.

Source ↗
2026-11-06

Arete-27B open weights (Nous Research)

Nous Research released Arete-27B, a supervised fine-tune of Alibaba's Qwen3.8-27B (Inkling as teacher) that targets state-aligned framing on politically sensitive topics. On Nous's own 803-prompt eval, answers with propaganda framing fell from 22% to 0.9% with no reported capability regression. Weights are on Hugging Face under Apache 2.0.

Source ↗
2026-11-06

HyperFlow H3 Flash open weights (Video Rebirth)

Video Rebirth released HyperFlow H3 Flash, a ~14B compressed, few-step version of MiniMax-H3 that generates 5–15 second video with synchronized stereo audio in 4–8 steps. INT8 and GGUF builds (Q4_K_M is 9.1 GB) run in ComfyUI on 24 GB+ GPUs. It ships under the MiniMax H3 Community License, which restricts territory (EU, UK, South Korea and US need MiniMax authorization).

Source ↗
2026-11-06

Kanana-2-v-decide open weights (Kakao)

Kakao released Kanana-2-v-decide-3b-2611, a 3.67B multimodal decision model that returns probabilities over choice, yes/no or score options instead of generating text. It accepts text, JSON or chat with up to 10 images (20,480-token context) and is built on Kanana-2-v-3b. Weights are on Hugging Face under the Kanana Open License.

Source ↗
2026-11-05

PLaMo 3 Translate open weights (Preferred Networks)

Preferred Networks released PLaMo 3 Translate weights on Hugging Face in four sizes (0.8B, 2B, 8B and 31B), with BF16, quantized, GGUF and MLX builds for on-device translation across 53 languages, plus context and glossary support. Licensed under the PLaMo Community License (free commercial use for organizations with revenue of JPY 1 billion or less, after registration).

Source ↗
2026-11-05

ERNIE-5.5-Flash-Base open weights (Baidu)

Baidu released ERNIE-5.5-Flash-Base, the end-of-pretraining checkpoint of a 280B sparse MoE (14B active) trained on 33.5T tokens with up to 128K training context and 1M-token fine-tuning/inference support. Weights (FP8 and BF16) are on Hugging Face under the ERNIE 5.5 Flash Base Model License; there is no mid- or post-training applied.

Source ↗
2026-11-04

Ripple-65B Preview open weights (Ricursive)

Ricursive Intelligence released Ripple-65B Preview, a 65B dense Transformer pretrained from scratch on 12T tokens with up to 1M context, aimed at agentic coding. Ricursive reports U.S. open-weight SOTA on Terminal-Bench 2.1 (80.52) ahead of much larger open models, trained for about $32M full R&D cost. Weights are on Hugging Face under Apache 2.0 (BF16 instruct + base, NVFP4 variant), with API access on Together AI, GMI Cloud and OpenRouter.

Source ↗
2026-11-04

Kimi K3.1 Pro open weights (Moonshot AI)

Moonshot AI released the full open weights of Kimi K3.1 Pro on Hugging Face as promised: a 2.8T-parameter multimodal MoE (104B active) with a 1M-token context, MoonViT-V2 vision, and MXFP4 weights / MXFP8 activations (~1.56 TB). It is natively agentic and multimodal (text + image). vLLM shipped day-0 support the same day (NVIDIA, AMD, TPU, Alibaba Zhenwu). License is the Kimi K3.1 Pro License (not MIT; large commercial/MaaS use has extra terms). API access continues on platform.kimi.ai; this entry is the weight drop (the Oct 23 API launch is already logged).

Source ↗
2026-11-03

Sarvam Prism open weights (Sarvam AI)

Sarvam AI released Prism, an open-weight decision model built as a LoRA fine-tune of Sarvam 105B. Instead of generating text, it takes a context, a question and a fixed set of options (yes/no, pick-one or a score) and returns a calibrated probability for every option in a single forward pass, for routing, triage, moderation, judging agent runs and prompt-injection checks. Sarvam reports a 71.8 average across three decision benchmarks with a 44 ms median response on a B200, and support for English, 10 Indian languages and Hinglish. Weights are on Hugging Face under Apache 2.0 (one FP8 GPU with about 120 GB of memory), and it is also available as a hosted Decisions API.

Source ↗
2026-11-02

Hydrogen-1.0 open weights (Deep Principle)

Deep Principle, a Hangzhou AI-for-science company, released Hydrogen-1.0, a 753B-parameter open-weight model for chemistry, materials science and life sciences, built by post-training Z.ai's GLM-5.3 on domain data, its own laboratory records and agent discovery trajectories. It is meant to help researchers reason through scientific questions, write and run simulation code, plan experiments and interpret results such as NMR spectra. Deep Principle reports 38.3% on FrontierScience-Research (5 points above GPT-6 Astra) and 93.0% on GPQA Diamond in its own evaluations. Weights are on Hugging Face and ModelScope under the MIT license (BF16, plus NVFP4 and MXFP4 builds), with a 1M-token context and up to 128K output tokens; the model is also available by API through NeuroSpark and inside Deep Principle's Mira AI Scientist.

Source ↗

October 2026

All 108 releases from October 2026 →

2026-10-30

Mage-VL-Cowork open weights (Microsoft)

Microsoft's Mage team released Mage-VL-Cowork, a 4B-parameter computer-use agent model that decides at each step whether to click through the screen (screenshot-grounded mouse and keyboard actions) or to write code and terminal commands, inside one policy. Microsoft reports 66.0% on OSWorld-Verified with a 50-step budget, ahead of much larger open agents such as Qwen3.6-27B (63.9%). The weights are on Hugging Face (microsoft/Mage-VL-Cowork) under Apache-2.0, with a technical report and training recipe on GitHub.

Source ↗
2026-10-30

Drex Max v1 open weights (Nace AI)

Nace AI open-sourced Drex Max v1, a dense 27B multimodal decision model. Instead of generating text, it reads a context (text, JSON or images) plus one or more typed questions and returns a probability for every option, up to 255, in a single forward pass, which suits jobs like ticket routing, yes/no checks and ratings. Nace reports 67.52 on the public Decision Index 0.4 and 64 on the Artificial Analysis Decision Index. Weights are on Hugging Face under the Nace.AI Open RAIL-M license in bf16, Q8_0 GGUF and affine Q4 formats; the Q4 build runs locally on Apple silicon Macs with 24 GB of memory (MLX or Nace's Kapitsa), and hosted inference is also available.

Source ↗
2026-10-29

Pika H3 Flex open weights (Pika)

Pika released H3 Flex, its post-trained version of MiniMax's H3 video model that generates in six steps instead of 50. Pika says it can make a 15-second clip with synchronized audio at 768p in as little as seven seconds of transformer compute, works from up to 12 references, exports up to five separate audio stems (dialogue, score, SFX, Foley, ambience) and finishes at up to 4K. The open release (generator adapters and inference code) covers text-to-video, first/last-frame and reference-to-video at 480p and 768p; the 1080p–4K upsampler and stem separation stay Pika-hosted. It is available now through Pika's API and comes to the Pika creation app on November 2. Pika says it ranked #1 among open-weight models (#7 overall) on DesignArena's text-to-video board at launch.

Source ↗
2026-10-29

Mistral Large 4 (Le Chonk) open weights

Mistral released the weights of an improved Mistral Large 4 ("Le Chonk"), its 1-trillion-parameter, natively multimodal mixture-of-experts model, which entered public preview earlier in October. After three more weeks of reinforcement learning, Mistral says it leads open-weight models on Cognition's FrontierCode 2.0 (19.7%) and Terminal-Bench 4.0 (41%), scores 48% on the Artificial Analysis Coding Agent Index (tied with GLM-5.3 while using 47% fewer tokens) and 51% on its Cyber Index, and now offers low, medium and high reasoning modes. Weights are on Hugging Face under the Mistral Ecosystem Open-Weight license (a modified MIT license; Model-as-a-Service businesses need a separate commercial license). The API costs $0.70 per million input tokens and $2.10 per million output tokens, and the model is also on OpenRouter, Microsoft Foundry and NVIDIA NIM.

Source ↗
2026-10-29

Apriel Voice 27B Preview open weights (ServiceNow)

ServiceNow AI Research released Apriel Voice 27B Preview, an open speech-language model that hears spoken requests directly and returns text responses and tool calls from one shared backbone, instead of chaining separate speech-to-text and dialogue models. It pairs a Whisper large-v3-turbo speech encoder with Google's Gemma 4 26B-A4B and keeps text and image input; speech output comes from a separate, swappable synthesizer. ServiceNow reports 65.51% task success on τ-Voice at 1.91 s mean latency and 98.90 on Big Bench Audio in thinking mode. Weights are on Hugging Face under the MIT license with vLLM serving support, and the Apriel Live web demo lets you talk to it.

Source ↗
2026-10-28

Replica open-source AI prototyping tool (GitHub)

GitHub Design open-sourced Replica (MIT license), an AI prototyping tool that turns natural-language ideas into interactive prototypes built from your team's own design-system components, patterns and docs. Designers set the guidance, and anyone on the team can create and refine prototypes from the GitHub Copilot app. Prototypes can be shared for comments on an optional, self-hosted Replica Hub.

Source ↗
2026-10-28

Clover-35B-A3B-Preview open weights (DiffuSpace)

DiffuSpace (founded by members of HKU NLP's Dream team) released Clover-35B-A3B-Preview, an open-weight diffusion language model that generates text in parallel blocks instead of one token at a time. It is a 35B-total, 3B-active MoE with a thinking mode and 128K context, reported at 680 tokens/s on a single GB300 GPU (BF16) and 73.6 on SWE-bench Verified. Weights are Apache-2.0 on Hugging Face and ModelScope, with SGLang and llama.cpp support.

Source ↗
2026-10-27

Ling-3.1-flash-1019 open weights (InclusionAI / Ant Group)

Ant Group's Ling team open-sourced Ling-3.1-flash-1019 under the MIT license. It is a post-trained upgrade of Ling-3.1-flash (560B total / 25B active hybrid-linear MoE) that scores 44 on the Artificial Analysis Intelligence Index, up from 41, with gains in agentic and scientific tasks. The model card claims a 1M-token context. Also released: Ling-3.1-flash-base, a BF16 base checkpoint for further training. The FP8 weights are about 574 GB.

Source ↗
2026-10-23

MiniMax M3.1 Flash open weights

MiniMax published the weights for M3.1 Flash (428B total, 23B active per token, 1M context, text/image/video input) on Hugging Face, four days after the Oct 19 API launch. Day-0 serving is on vLLM and Together AI. It is released under the MiniMax community license (attribution required; prior authorization above $20M annual revenue), not MIT/Apache.

Source ↗
2026-10-23

Hi3D V1.0 open source (formerly Sparc3D)

Hi3D open-sourced Hi3D V1.0 (formerly Sparc3D), an image-to-3D model trained on about 2M curated 3D assets with sparse-grid geometry reconstruction up to 1536³. Code is on GitHub and weights are on Hugging Face, with a free online demo. Hi3D also cut credit use on its hosted platform by over 50%.

Source ↗

Frequently asked questions

What is the latest open-weight AI model release?

Hy4 open weights (Tencent) (2026-11-20). Tencent published the open weights for Hy4, the natively multimodal flagship mixture-of-experts model it launched by API on Nov 4: 770B total parameters with 49B active per token (plus a 1.23B vision encoder and a 3.3B DFly2 speculative-decoding drafter), 78 layers, a 1M-token context window, text, image and video understanding, and computer use.

How many open-weight AI models were released in November 2026?

SIloadout has verified 21 open-weight or open-source AI releases in November 2026 so far (through November 20), and 33 in total since October 16, 2026.

Does open weights mean open source?

Not always. Open weights means the trained model files can be downloaded and run yourself, but the license can still restrict use (for example non-commercial, community or custom licenses). Each entry's summary names the license when the vendor states it; check the model card before commercial use.

How does SIloadout verify releases?

Each launch is checked against the vendor's own post, its official X account or credible press before it is added, and every source link is confirmed to load. Rumored, unreleased or ambiguously dated items are left out.

All releases · Atom feed · JSON Feed · CSV

Browse by month: November 2026 (83) · October 2026 (108)

Open models only: New open-weight and open-source AI releases (33)

By company: Google (10) · OpenAI (7) · Microsoft (6) · Perplexity (5) · Tencent (5) · Anthropic (4) · ElevenLabs (4) · NVIDIA (4) · Baidu (3) · GitHub (3) · HeyGen (3) · Meta (3) · MiniMax (3) · Mistral AI (3) · Moonshot AI (Kimi) (3) · All companies