October 2026 AI Model Releases for Creators: FLUX 3 Klein, Vidu Q4 and Suno Speech
Key takeaways
FLUX 3 Image Klein (October 15) is the month's most useful release for anyone generating images locally: 5.6 billion parameters, up to 4K output, 10 reference images and free commercial use for companies under $5 million in revenue.
FLUX 3 Image (October 1) is Black Forest Labs' closed flagship for generating and editing in one model, with bounding-box placement in the prompt and native 2K and 4K output.
Vidu Q4 Preview takes up to 15 reference images and three voice clips and renders up to 16 seconds at 4K, but only from images; there's no text-to-video in the preview.
Kandinsky 6.0 Video is open weights under the MIT license and generates about 5 seconds of video with synced audio, including lip sync.
Suno Speech, in open beta since October 1, makes a spoken voice track and background music together from a script.
The October releases at a glance
Release | Maker | Date | Type | Access |
|---|---|---|---|---|
FLUX 3 Image | Black Forest Labs | October 1 (update October 22) | Image generation and editing | API and Playground, $0.041 to $0.607 per image |
Suno Speech (beta) | Suno | October 1 | Voice with background music | Suno app |
Kandinsky 6.0 Video | Kandinsky Lab | October 6 | Video with audio | Open weights, MIT license |
Vidu Q4 Preview | Shengshu | Early October | Video from images and references | Vidu app and API, from $0.045 per second |
Odyssey-3 | Odyssey | October 8 | Interactive world model | Research preview, free Flash tier |
Qwen-Image-2.1-Turbo | Alibaba's Qwen team | Early October | Fast image generation and editing | Open weights, non-commercial license |
FLUX 3 Image Klein | Black Forest Labs | October 15 | Image generation and editing | Open weights and API, from $0.008 per image |
VideoGen Model 1 | VideoGen | Announced October 20 | Long multi-shot video | Selected partners only |
FLUX 3 Image Klein: the open model most people can run
Klein is a 5.6-billion-parameter version of FLUX 3 Image with open weights on Hugging Face. It generates, edits and combines up to 10 reference images, with native output up to 4K, and comes as a base model for fine-tuning and a step-distilled model for speed. The INT8 build runs on an 8 GB RTX card, and Diffusers supported it from day one.
Check the license before commercial use. The FLUX Kommunity License allows free commercial use for individuals and companies under $5 million in annual revenue; larger businesses need a self-hosting license or the API, which charges $0.008 to $0.036 an image.
FLUX 3 Image: one model for making and editing
FLUX 3 Image, released October 1, merges generation and editing behind a single endpoint. You can place objects by writing bounding boxes into the prompt, make pixel-exact local edits and pass up to 10 references, with native output at 2K or 4K. An October 22 update added exact custom output sizes, an option to match the input size when editing and better results on single-reference edits, at the same prices. It's closed: access is through BFL's API and Playground, at $0.041 to $0.607 an image.
Vidu Q4 Preview: 4K and many references
Shengshu's new flagship renders up to 16 seconds at up to 4K with 10-bit color. It accepts up to 15 reference images for characters, products, locations and styles, plus up to three voice clips to keep a character sounding the same across shots. The preview works only from images and references, not from text. API pricing runs from $0.045 a second at 540p to $0.39 a second at 4K, with image- and reference-to-video 30% off through November 30.
Kandinsky 6.0 Video: open weights with sound
Kandinsky Lab, part of the Russian bank Sber, released Kandinsky 6.0 Video in Pro (29 billion parameters) and Lite (3 billion) versions. It makes clips of about 5 seconds at 24 fps with synced 44 kHz audio covering speech with lip sync, ambience and music, or silent output if you prefer. A separate 1.4-billion-parameter model upscales to 1080p. The MIT license allows commercial use.
Suno Speech: voice and music in one track
Suno Speech generates a spoken track and an original music bed together from an idea or a script, which suits podcast intros, ads and narrated social clips. It opened to everyone on October 1 after about a month of testing. Suno itself says accents can drift and pauses can run long, and it hasn't published separate pricing or an API for Speech.
Qwen-Image-2.1-Turbo: fast, but research only
Alibaba's Qwen team released a 7-billion-parameter Turbo version of Qwen-Image-2.1 that generates and edits in eight steps instead of 40, at up to 2048 by 2048 pixels or 2752 by 1536. Weights are on Hugging Face. The Qwen Research License does not allow commercial use, so keep it for experiments; client work needs a commercially licensed model.
Odyssey-3 and VideoGen Model 1: worth watching
Odyssey-3 is a world model: it builds an environment you can move through in real time, in first or third person. Its October 8 research preview has a free Flash tier but no API or pricing, and Odyssey pitches it mainly at robotics and AI training.
VideoGen announced Model 1 on October 20. It assembles multi-shot videos up to 10 minutes long, but only selected partners can use it so far, and there's no public pricing or outside testing yet.
Released just before October
Three September releases landed close enough to matter here:
Suno v6, v6-wild and v6-mini (September 9)
2K and 4K output for FLUX 3 Video (September 10), at $0.40 to $0.95 a second
The Qwen-Image-2.1 base model, in late September, with RGBA output and 10 references
Local models and more references
Klein's 8 GB requirement puts a current, commercially usable image model on a mid-range gaming card, and Kandinsky 6.0 does the same for short video with sound, though its Pro model needs far more memory. FLUX 3 Image and Klein take 10 references and Vidu Q4 takes 15 plus voice clips, so a character or product can stay consistent across a series.
Frequently asked questions
What AI models were released in October 2026?
October's creative releases included FLUX 3 Image and FLUX 3 Image Klein from Black Forest Labs, Vidu Q4 Preview from Shengshu, Kandinsky 6.0 Video, Suno Speech in open beta, Qwen-Image-2.1-Turbo and Odyssey-3's research preview. VideoGen announced Model 1 for selected partners.
Can I run FLUX 3 Klein on my own computer?
Yes. The INT8 build of FLUX 3 Image Klein runs on an 8 GB RTX graphics card, and the weights are free on Hugging Face. Commercial use is free for individuals and companies under $5 million in annual revenue.
Which new video models generate sound?
Kandinsky 6.0 Video generates synced audio, including speech with lip sync, ambience and music. Vidu Q4 Preview uses voice reference clips to keep a character's voice consistent.
Is Vidu Q4 available to everyone?
Yes, as a preview through Vidu's app and API. It works from images and references only, and API prices run from $0.045 a second at 540p to $0.39 a second at 4K.
Which October models are open source?
FLUX 3 Image Klein, Kandinsky 6.0 Video and Qwen-Image-2.1-Turbo all have downloadable weights. Kandinsky uses the MIT license, Klein allows free commercial use under $5 million in revenue and Qwen's Turbo model is for non-commercial research only.