whatmodel.sh / image / FLUX.1 Kontext [dev]
FLUX.1 Kontext [dev]
↗ share
+ contribute
FLUX.1 Kontext [dev] is an open-weights image model from Black Forest Labs, released June 2025. Its published size is 12B parameters. Licence: flux-1-dev-non-commercial-license.
Model information sources
Release date ↗ · Parameter count ↗
12B = the rectified flow transformer as Black Forest Labs states it on the card; the text encoders and VAE used by the pipeline are not included (same convention as flux-1-dev).
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of FLUX.1 Kontext [dev], out of 197 with a score, then more from Black Forest Labs.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 124 |  DeepGen 1.0 | 45±3 | comparison only | ||
| 125 |  DALL·E 3 | 44±1 | comparison only | ||
| 126 | FLUX.1 Kontext [dev] | 44±1 | api $0.025/image +local | ||
| 127 | FLUX.1 [dev] | 44±1 | api $0.025/image +local | ||
| 128 |  UniPic2-SD3.5M-Kontext-2B | 44±6 | comparison only | ||
| more from Black Forest Labs | |||||
| 32 | FLUX.2 [max] | 64±1 | api $0.070/image | ||
| 36 | FLUX.2 [pro] | 63±2 | api $0.030/image | ||
| 37 | FLUX.2 [flex] | 62±2 | api $0.050/image | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #88 of 126 * · 27% of its score · 1 test
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ) here, though they rank higher overall. Trails Playground v2.5, Playground v2 (1024px aesthetic), Show-o2 7B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena text-to-image | 941.1 ✎ | ▃▂▃▄▇█▄▂▄▂·▂ beats 2/75 | 27% |
image to image / editing▁▁·▃▃▃█▃▃▄▁▁ #52 of 62 · 57% of its score · 9 tests
Beats FLUX1.1 [pro] here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Edit-Compass (EN, overall) | 38.6% ✎ | ▃▃▂·▃▆▂█·▃·▃ beats 9/20 | 15% |
| Arena image edit | 1149.2 ✎ | ▂▁▁▃▆▇█▄▆▅▁▂ beats 6/53 | 10% |
| CEdit-Bench-CN overall (G_O, GPT-4o judge) | 15.1% ✎ | ▃·········█▂ beats 1/7 | 6.9% |
| CEdit-Bench-EN overall (G_O, GPT-4o judge) | 59.3% ✎ | ▄·▄···▄██··▄ beats 0/7 | 6.1% |
| UnicBench (EN, overall) | 68.0% ✎ | ▂····▂▄▃▄·█▂ beats 4/11 | 5.8% |
| GEditBench v2 (overall Elo) | 869.0 ✎ | ▃▂····▂█▂▃▂▃ beats 1/14 | 4.0% |
| CREval-Bench creative image editing (Overall, GPT-4o QA evaluator) | 74.8% ✎ | ▄█▄··█··▄·▄▄ beats 4/8 | 3.4% |
| MultiTextEdit cross-lingual text-in-image editing (SemAvg, LVM judge) | 48.0% ✎ | ▄·▄▄▄▄··█▄·▄ beats 4/8 | 3.0% |
1 more test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ImagenWorld text-guided editing (human overall) | 70.0% ✎ | █··█··██···█ beats 3/4 | 2.5% |
other tests5 not in a task · 16% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| RealPref-Bench image editing (human user-study score) | 66.9% ✎ | █▄▄····▄···▄ beats 1/5 | 6.1% |
| WEAVEBench in-context multi-turn editing (Avg) | 68.9% ✎ | █▄▄·▄▄▄▄█·▄█ beats 12/12 | 3.7% |
| EdiVal multi-turn editing overall (EdiVal-O after turn 3) | 38.7% ✎ | ▂·▂▃·▂·▃▂·▃█ beats 5/14 | 3.5% |
| ViGoR-Bench zero-shot visual reasoning, image models (Result Avg) | 38.9% ✎ | ▂···▂·▂█▆▅·▂ beats 11/15 | 1.3% |
| PaintBench precise editing (mIoU, 20-task average) | 3.9% ✎ | ▃·····█▅▃·▃▅ beats 3/9 | 1.3% |
$ where to run it
Through an API
api · Replicate · $0.025/image · replicate.com ↗ · price sourceUSD per output image on Replicate's official black-forest-labs/flux-kontext-dev model (billing: $0.025 per output image, "or 40 images for $1"; hardware time not added). The current version is served from replicate-pruna/flux-kontext-dev (Pruna-optimized deployment). Product: Replicate model API. API model: black-forest-labs/flux-kontext-dev. Settings: one output image. Price basis: published list rate. fal's fal-ai/flux-kontext/dev bills $0.025 per megapixel, rounded up to the nearest megapixel. Currency: USD. Not a BFL direct tariff. Replicate official model page, observed 2026-09-26.
Other ways to run it
version
- Run it as published · needs ~21.3 GB · from github.com/leejet
sd-cli -r input.png --diffusion-model models/flux1-kontext-dev-Q4_K_M.gguf --vae models/ae.safetensors --clip_l models/clip_l.safetensors --t5xxl models/t5xxl_fp16.safetensors -p "change 'flux.cpp' to 'kontext.cpp'" --cfg-scale 1.0 --sampling-method euler -v --clip-on-cpu
Official weights and documentation: huggingface.co/black-forest-labs ↗
$ how the score was worked out · LMArena leaderboard dataset 37%, Edit-Compass & EditReward-Com… 15%, Meituan LongCat — LongCat-Ima… 13% and 10 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json