news

Black Forest Labs ships FLUX 3 Image Klein at 5.6B parameters, with every first-place claim scored in-house

The October 15 open weights release reports a 91 Elo gain over FLUX.2 klein in text-to-image while using 38% fewer parameters, and the comparison set was chosen by the company making the claim.

By Priya Shenoy · 18 October 2026

Black Forest Labs released FLUX 3 Image Klein on October 15 with open weights, 5.6 billion parameters, and a claim that it leads every other open weights image model in text-to-image, multi-reference generation and image editing. All three placements come from the company's own internal evaluations, against a comparison set the company selected.

A 5.6B model that keeps the 4K and bounding box features of its larger sibling

The release is a family rather than a single checkpoint. There is a base model intended for fine-tuning, a step distilled model for low latency generation, and FP8, INT8 and NVFP4 builds developed with NVIDIA support, all under the FLUX Kommunity License. The generator weights require 11.2 GB of memory in BF16 and 5.6 GB in FP8, and with the INT8 build the model runs on GeForce RTX cards with as little as 8 GB of VRAM through ComfyUI or Diffusers. Klein does text-to-image, editing and multi-reference generation with up to 10 reference images in one model, retains native 4K output from the larger FLUX 3 Image model, and accepts bounding box layouts for placing, moving, recoloring or removing elements.

The larger FLUX 3 model is already on our price radar, which gives some sense of what the paid end of this family costs per generation.

The ranking numbers are Black Forest Labs scoring Black Forest Labs

In internal evaluations the company places Klein first in text-to-image, multi-reference generation and image editing, ahead of larger open weights models and above the two closed models included in the comparison, Ideogram V4.5 and FLUX.2 pro. The parameter gap is the detail the announcement keeps returning to: 5.6B against 7B for Qwen-Image 2.1 and 12.9B for Krea 2. Measured against the company's own FLUX.2 klein at 9B, BFL reports a gain of 91 Elo in text-to-image and 70 in image editing while using 38% fewer parameters.

An Elo delta only carries information when the setup around it is visible: who or what voted, how many pairwise comparisons were collected, which prompts were used, and which of the five released builds was scored. The launch announcement does not attach those details to the figures, though BFL says a full technical report ships alongside the weights. Self-scored firsts are not unusual in image releases this autumn, and we have seen the same pattern from Ideogram and Tencent within the last month.

Text rendering is one of the capabilities Klein inherits a claim on, and FLUX 3 sits in our own benchmark for that category.

Free commercial use ends at $5 million in annual revenue

The licence is the detail with the most consequence for anyone planning to build on this. VentureBeat reports that the FLUX Kommunity License permits free commercial use for individuals and businesses whose annual revenue falls below $5 million, and that larger organizations need a separate arrangement for commercial self-hosting or can use the BFL API, where Klein starts at $0.009 for a 1K resolution image. Downloadable weights and unrestricted commercial rights are two different things here, and the revenue line is where they separate.

Day zero support landed in Diffusers, ComfyUI, SGLang and Ostris AI Toolkit, and the model is live on the BFL API and on fal. That makes the independent check easy to run: until a third-party arena posts a rating for Klein, the open weights lead remains a vendor claim with a vendor scoreboard behind it.

Sources: Black Forest Labs, GlobeNewswire, VentureBeat, The Decoder.

Priya Shenoy: Tracks what AI video actually costs across the platforms that resell access to the same handful of models. Treats a pricing page as a claim, not a fact, until someone checks it.