News
New releases and license changes, with the numbers checked and the sources linked.
- Apertus 1.5: Switzerland's fully open model reads images and audio
ETH Zurich, EPFL and CSCS released Apertus 1.5 on 24 July: fully open 8B and 70B models with image and audio input, a 262K context and Apache 2.0. Audio understanding is experimental.
- Clef: Cloudflare's open decision models for agent workflows
Cloudflare released Clef and Clef-flash under Apache 2.0: 27B and 9B decision models, post-trained from Qwen, that return a probability for each option you define instead of writing text.
- Cosmos 3: NVIDIA's open world models for robots and driving
NVIDIA released Cosmos 3 under OpenMDW-1.1: 4B, 16B and 64B world models that generate video, images and robot actions and reason about video, for robotics and autonomous driving.
- EmbeddingGemma 2 embeds text, images, video and audio under Apache 2.0
Google DeepMind's 740M EmbeddingGemma 2 maps text, code, images, video and audio into one 768-dimension space, runs on-device and moves from the Gemma license to Apache 2.0.
- Baidu's ERNIE 5.5 Flash Base: 280B parameters, license excludes the EU
Baidu released ERNIE-5.5-Flash-Base on 5 November: a 280B mixture-of-experts base model with 14B active and a 1M context. Its license excludes the EU, the UK and South Korea.
- Four open translation models, four different yardsticks
TranslateGemma, Hy-MT2, Index-Translate and PLaMo 3 Translate each report quality on different test sets, metrics and rivals, so their claims cannot be compared directly.
- Free until you grow: where open licenses draw the line
Thirteen licenses in our catalog end free commercial use at a revenue or user threshold, from US$500,000 to US$50 million. Here is how the lines, the counting and the duties differ.
- Tencent's Hy4 preview: 770B parameters, 1M context and Apache 2.0
Tencent's Hy4 preview is a 770B mixture-of-experts model with 49B active and a 1M-token context, under Apache 2.0 after Hy3's preview had territory limits. Tencent calls it an early version.
- Hydrogen-1.0: a 753B open science model built on GLM-5.3
Deep Principle released Hydrogen-1.0, a 753B science model post-trained from GLM-5.3 under an MIT label, with 1M context and 4-bit builds. All benchmark claims are its own.
- Index-Translate: Bilibili's open translators for 150 languages
Bilibili's Index team released Index-Translate, Apache 2.0 translation models in 2B, 9B and 35B-A3B sizes for 150 languages, with image translation and glossary, format and tone controls.
- Korea's four sovereign AI models: open weights, three teams advance
Four open-weight models of 250B parameters or more came out of South Korea's sovereign AI project. Compare their sizes, licenses and hardware needs, and see which teams advanced.
- Krea 2: an open image model trained from scratch for variety
Krea released Krea 2, a 12.8B text-to-image model trained from scratch, as a Raw base and an 8-step Turbo. Commercial use is free only below US$1 million in annual revenue.
- Meituan's LongCat-2.5: 1.6T parameters, vision and the plain MIT license
Meituan released LongCat-2.5 as open weights on 9 November: a 1.6T mixture of experts with 48B active, image and video input, a 1M-token context claim and the MIT license.
- MiniMax H3-Image: 16-megapixel open weights, but not for the US or EU
MiniMax's 33B H3-Image makes 16-megapixel pictures in six steps and edits with up to 16 references. Its license excludes the US, EU, UK and South Korea and caps commercial use at US$1M.
- MOSS-Image: a 5B open image model for long text and JSON layouts
OpenMOSS released MOSS-Image under Apache 2.0: a 5B generator that renders long English and Chinese text and follows JSON layouts, with a low-memory profile near 4 GB of VRAM.
- Nanbeige4.3-33B: a 33B agent model with a 23 GB 4-bit build
BOSS Zhipin's Nanbeige lab published Nanbeige4.3-33B on 24 October: a dense 33B text model with a 512K context, an INT4 build of 22.6 GB and an Apache 2.0 tag but no LICENSE file.
- NVIDIA's Nemotron 3.5 Embed and Parse: document pages as images
NVIDIA released Nemotron 3.5 Embed, which searches page images and text without OCR, and Nemotron 3.5 Parse, a 631M page parser, both under OpenMDW-1.1 for commercial use.
- Open models that listen: nine take audio, but only three answer in speech
Gemma 4, Apertus 1.5, Inkling, MiMo-V2.6, Muse Spark Open, Nemotron 3 Nano Omni, Qwen3-Omni, MiniCPM-o 4.5 and LFM3-Omni compared by audio input, speech output, size and license.
- Open music models grew up in 2026: seven models for songs, instruments and sound effects
Seven open music and sound models arrived between January and September 2026, from song writers such as ACE-Step 1.5 to a live instrument from Google. Compare jobs, hardware and licenses.
- Eleven open OCR models in a year: small and hard to compare
Eleven open OCR models arrived between October 2025 and October 2026. Most have 1B to 5B parameters and Apache 2.0 or MIT licenses, but benchmark versions differ, so their scores cannot be ranked.
- Open speech models: transcription is open, text-to-speech is mixed
Ten of our eleven open speech-to-text models have no commercial limit, but only six of fourteen text-to-speech models are plain Apache 2.0. The rest are non-commercial, capped, restricted or retired.
- PLaMo 3 Translate: PFN's open translators for 53 languages
Preferred Networks released PLaMo 3 Translate in 0.8B to 31B sizes for 53 languages, with GGUF files from 0.57 GB and a license that limits commercial use to 1 billion yen of revenue.
- Raon-1: KRAFTON's Korean-first 28.8B model and a 3.1B on-device sibling
KRAFTON AI released Raon-1 mini preview, a 28.8B mixture of experts trained from scratch, and the 3.1B Raon-1 nano distilled from it, both under Apache 2.0 with a 262K context.
- Ripple-65B Preview: a dense 65B agent model under Apache 2.0
Ricursive Intelligence, a chip-design lab, released Ripple-65B Preview on 4 November: a dense 65B text model for agentic coding with 1M context, Apache 2.0 and a 57 GB NVFP4 build.
- Sber's open models: a 432B chat model, speech and embeddings
Sber's GigaChat team has five MIT-licensed entries on Hugging Face: GigaChat 3.5 Ultra, 3.1 Lightning, GigaAM-v3, GigaAM Multilingual and Giga-Embeddings. Sizes, scores and caveats.
- PrismML's Ternary Bonsai 2: a 27B-class model in 5.95 GB
PrismML released Ternary Bonsai 2: Qwen models stored with three-value weights, from a 27B model in 5.95 GB to a 2B in 0.43 GB, with image input, a 262K context and Apache 2.0.
- Tiny text-to-speech models that run on a CPU: four open options compared
Kokoro, Pocket TTS, Phonon 31M and Supertonic 3 are open text-to-speech models of 100M parameters or fewer that run on a laptop CPU. Compare sizes, languages, voice cloning and licenses.
- ZONOS3-Mini: open TTS in 43 languages that runs in a browser
Zyphra released ZONOS3-Mini on 9 November: an 800M mixture-of-experts text-to-speech model with 100M active parameters, 43 languages, voice cloning and Apache 2.0 weights.
- Giant open models now ship in 4-bit: what it means for memory planning
Eight models in our catalog ship their main checkpoint in a 4-bit-class format, and nineteen more add official 4-bit builds. See the file sizes and how they change the hardware you need.
- Open licenses: free to build on, but hosting needs a deal
Six of the biggest 2026 open-weights licenses let you build products freely but require a separate agreement or review from businesses that sell hosted access. Here is how the thresholds differ.
- Qwen3.8: three language models, three licenses, and a fourth for images
Alibaba's Qwen3.8 models ship under Apache 2.0, the Qwen3.8-Max License and Qwen Community 1.0, while Qwen-Image-2.1 is research-only. A map of what each license lets you do.
- Moonshot releases Kimi K3.1 Pro: 2.8T parameters, open weights
Moonshot AI published Kimi K3.1 Pro on 2 November 2026: a 2.8T mixture-of-experts model for long coding and office tasks, with a 1M-token context and 4-bit weights of about 1.56 TB.
- Mistral Large 4 goes open-weights: 1T parameters, trained in Europe
Mistral published the weights of Mistral Large 4 on 29 October 2026: a multimodal mixture-of-experts model with a 1M-token context, 160+ languages and a license with a hosting condition.
- NVIDIA releases Nemotron 3.5 Transcribe, a promptable speech model
NVIDIA published Nemotron 3.5 Transcribe on 28 October 2026 under OpenMDW-1.1: 40 languages, speaker labels, translation and prompt-controlled transcript style, for recordings up to an hour.
- Liquid AI announces LFM3: six small models, text to voice
Liquid AI released the LFM3 family on 27 October 2026: text, vision, audio and omni models of about 2B parameters built for devices, plus LFM3-Audio-2B for voice, under the LFM Open License.
- MiniMax releases M3.1 Flash, a 428B model built for 4-bit
MiniMax published the M3.1 Flash weights on 17 October 2026: 428B parameters, 23B active, a 1M-token context and a 252 GB checkpoint, under a community license with credit and notice conditions.
- Black Forest Labs releases FLUX 3 Image Klein with layout control
Black Forest Labs published FLUX 3 Image Klein on 15 October 2026: 5.6B parameters, editing with up to 10 references, box-based layout control, an 8 GB INT8 build and a revenue-capped license.
- Meta releases Muse Spark Open, an 816B open-weights model
Meta published Muse Spark Open on 15 October 2026: 816B parameters, 42B active, text, image, video and audio input, a 1M-token context, a gated download and a 'Built with' credit requirement.