whatmodel.sh / image / Step1X-Edit v1.0
Step1X-Edit v1.0
↗ share
+ contribute
Step1X-Edit v1.0 is an open-weights image model from StepFun, released April 2025. Its published size is 12.43B parameters. Licence: apache-2.0.
Catalogue identity notes
StepFun's original Step1X-Edit, released 25 Apr 2025 (GitHub README news: 'We release the inference code and model weights of Step1X-Edit'); checkpoint step1x-edit-i1258.safetensors in stepfun-ai/Step1X-Edit, which since 9 Jul 2025 also holds the v1.1 file step1x-edit-v1p1-official.safetensors (catalogued separately as step1x-edit-v1-1). StepFun's own README table labels it 'Step1X-Edit (v1.0)'; EdiVal Table 3 lists 'Step1X-Edit' with release-date column 25.04.25 and weights 'https://huggingface.co/stepfun-ai/Step1X-Edit' (Appendix M.5), and the EdiVal ledger scope already identifies that row as the v1.0 repo. No local variant: the official route (inference.py --version v1.0) needs flash-attn built for the local CUDA/torch and a manually placed Qwen2.5-VL-7B-Instruct folder, and the README gives no single copy-paste command; peak memory per the README A2 table is 49.8 GB at 1024 (18 GB with --quantized --offload).
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Diffusion transformer only: OUR COUNT from the safetensors header of step1x-edit-i1258.safetensors (the v1.0 checkpoint; inference.py maps --version v1.0 to this file) at the pinned revision: 12,427,800,385 BF16. The Qwen2.5-VL-7B-Instruct encoder and vae.safetensors are not included. Observed 2026-10-09.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Step1X-Edit v1.0, out of 197 with a score, then more from StepFun.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 153 |  Janus-Pro-7B | 38±2 | api $0.0020/image +local | ||
| 154 | NextStep-1-Large-Edit | 37±3 | comparison only | ||
| 155 | Step1X-Edit v1.0 | 37±4 | comparison only | ||
| 156 |  SDXL-Lightning (4-step UNet) | 37±3 | local | ||
| 157 |  Kandinsky 2.2 | 37±5 | local | ||
| more from StepFun | |||||
| 89 | Step1X-Edit v1.2 | 53±2 | local | ||
| 103 | Step1X-Edit v1.2 Preview | 50±5 | comparison only | ||
| 108 | Step1X-Edit v1.1 | 48±3 | local | ||
$ what it's good at
image to image / editing▁▁▃▃▂▃█▃▃▃▁▁ #77 of 81 * · 72% of its score · 2 tests
Trails OmniGen (v1) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena image edit | 998.3 ✎ | ▃▁▁▂▆▇█▄▆▅▁▂ beats 0/53 | 72% |
| GEdit-Bench-EN overall (G_O, GPT-4.1 judge) lab | 64.4% ✎ | ▄·····█·▄▄·▄ beats 1/5 | — |
other tests1 not in a task · 23% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| EdiVal multi-turn editing overall (EdiVal-O after turn 3) | 39.0% ✎ | ▂·▂▃·▂·▃▂·▃█ beats 6/14 | 23% |
$ where to run it
No verified API or download recorded yet · add one
Weights and documentation: huggingface.co/stepfun-ai ↗
$ how the score was worked out · LMArena leaderboard dataset 72%, EdiVal-Agent: An Object-Centr… 23%
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
| source | results | share of the score |
|---|---|---|
 LMArena leaderboard dataset | 1 | 72% |
 EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (arXiv:2509.13399v3) | 1 | 23% |
| prior · pull toward the middle | 5% | |
| weighted average | = 37.1 |
not counted (1)
| stepfun-ai/Step1X-Edit repository README results | GEdit-Bench-EN overall (G_O, GPT-4.1 judge) [correct] | 64.4% | fewer than three models have this test yet, or this source ran it for this model alone |
$ 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