![Flux (FLUX 3 + FLUX.2 [klein]) logo](/logos/flux.png){kind=small}
Flux (FLUX 3 + FLUX.2 [klein]) Pricing
All plans and pricing as of 2026-09-28
FLUX.2 [klein] (Open-core)
- ✓Launched 2026-01-15 -- 4B + 9B parameter variants
- ✓Apache 2.0 on open-core weights
- ✓Under 0.5s generation on modern hardware, 4MP coherence
- ✓Multi-reference mode + native image editing
- ✓Requires 13 GB VRAM minimum (4B variant)
FLUX 3 Video API
- ✓Text-to-video and image-to-video: $0.17 (HD) / $0.29 (Full HD) / $0.40 (QHD) / $0.80 (4K UHD) per second, clips of 5-20 seconds (docs.bfl.ai, 2026-09-28)
- ✓Video continuation from your own clip: $0.41-$0.95 per second, 5-15 seconds; draft renders $0.06/s (t2v, i2v) and $0.12/s (v2v)
- ✓Native audio, dialogue and lip-sync in many languages, multiple shots and camera angles in one generation, 24 fps up to 3840 x 2176
- ✓Docs still label FLUX 3 a preview model; Omni Reference with images and videos is promised but not live
- ✓FLUX Video Upscale tool (2026-08-20): regenerate 480p+ video at up to 4K, Precise $0.07 or Creative $0.10 per megapixel-second
FLUX 3 Action (open weights)
- ✓7B world action model released 2026-09-22 on Hugging Face (base weights plus SO-101 and DROID robot policies), runs through LeRobot
- ✓FLUX Kommunity License: non-commercial and non-production use, or commercial use only if your gross annual revenue is under $5M; a paid licence otherwise
- ✓Vendor claim, unreplicated: new state of the art on the RoboLab-120 leaderboard at under half the parameters of the previous best open model, up to 3.95x faster
- ✓Text encoder is an unmodified Qwen3-VL-4B-Instruct; joint action plus roughly two seconds of predicted video per step
Flux.1 Schnell (legacy)
- ✓Apache 2.0 license
- ✓Run locally
- ✓Fast generation
- ✓Superseded by FLUX.2 [klein] for most use cases
FLUX.2 [max] / [pro] API
- ✓BFL's hosted API -- highest quality tier
- ✓Multi-reference + editing + higher resolution than [klein]
- ✓Commercial license included
- ✓Positioned vs Nano Banana 2 and Midjourney
Is Flux (FLUX 3 + FLUX.2 [klein]) Worth the Price?
Value Score: 8.5/10
Overall Score: 7.8/10 · Technically savvy users who want the best possible image quality and are willing to set up local inference. Also great for developers who want an open-source model they can fine-tune and deploy on their own infrastructure.
Flux earned its reputation as the photorealism benchmark in open image generation, and FLUX.2 [klein] is still the model to run locally if you have 13 GB of VRAM and the patience for ComfyUI. What changed in mid-2026 is scope: FLUX 3 makes Black Forest Labs a video-with-audio vendor with a flat per-second API price, and FLUX 3 Action points the same model at robots. Neither is plug-and-play -- the video side is API-only and still labelled a preview, and the Action weights carry a revenue-capped licence. If you want the best open image weights and are willing to invest in setup, this is still the answer. If you want to type a prompt and get a picture, look elsewhere; if you want 20-second clips with sound on a metered API, FLUX 3 Video is now a real option to price against Veo and Kling.
The Tier List Tuesday
Weekly newsletter: tier movers, new entrants, and the VS of the week. Built from our daily AI-tool sweeps. No spam, unsubscribe anytime.
How Flux (FLUX 3 + FLUX.2 [klein]) Pricing Compares
| Tool | Free Tier | Starting Price | Value Score | Overall |
|---|---|---|---|---|
| Flux (FLUX 3 + FLUX.2 [klein])(this tool) | Yes | $0 | 8.5/10 | 7.8 |
| Nano Banana 2 (Gemini 3.1 Flash Image) | Yes | $0 | 8.5/10 | 8.9 |
| Meta Muse (Image + Video) | Yes | $0 | 9/10 | 8.4 |
| Leonardo AI | Yes | $0 | 9/10 | 8.3 |
| Stable Diffusion | Yes | $0 | 10/10 | 8.0 |
| Midjourney | No | $10/month | 7/10 | 7.8 |