Up to date
stepfun-ai Apache 2.0 out 9 months ago
10.2B parameters, images and text in, text out, 64K tokens of context.
Verified 2026-10-25 Rank #638 in LLMs · score 73.2 method v1 Sources 3 Downloads 30d 10.2k · #505 of 619 tracked
Picked by rule from 10 tracked pieces: the lab's own post, the most shared, a field report, the latest.
The 1 pick above, plus these 9. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 9 quantized builds and 2 Spaces.
Top Spaces running it: stepfun-ai/Step3-VL-10B ♥ 25
Weights only, an estimate: leave room for the KV cache. Method
About 6.1 GB at INT4, 12 GB at FP8, 24 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-01-13), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
13 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | image-text-to-text |
|---|---|
| Architectures | StepVLForConditionalGeneration |
| Model type | step_robotics |
| Dtype | bfloat16 |
| Params by dtype | BF16: 10,171,750,144 |
| Size band | L |
| MoE | no |
| Context length | 65,536 (config.json max_position_embeddings) |
| Papers | arXiv:2601.09668 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · StepVLForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · StepVLForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for Step3-VL-10B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 24 GB · Single high-end GPU (<80 GB) |
|---|---|
| FP8 (1 byte/param) | 12 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 6.1 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
Base model: stepfun-ai/Step3-VL-10B-Base
quantized: 9finetune: 4
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| seanbailey518/Step3-VL-10B-GGUF | quantized | GGUF | — | — | other | seanbailey518 | 11 | 21.5k |
| QuantTrio/Step3-VL-10B-AWQ | quantized | AWQ | 10.2B | L | apache-2.0 | QuantTrio | 5 | 13.5k |
| huihui-ai/Huihui-Step3-VL-10B-abliterated | finetune | — | 10.2B | L | apache-2.0 | huihui-ai | 20 | 3k |
| cyankiwi/Step3-VL-10B-AWQ-4bit | quantized | AWQ | 4.31B | M | apache-2.0 | cyankiwi | 0 | 3.4k |
| np-deploys/Step3-VL-10B-AWQ-4bit | quantized | AWQ | 4.31B | M | apache-2.0 | np-deploys | 0 | 2.7k |
| JamePeng2023/Step3-VL-10B-GGUF | quantized | GGUF | — | — | apache-2.0 | JamePeng2023 | 0 | 2.4k |
| Vastined/Step3-VL-10B-GGUF | quantized | GGUF | — | — | apache-2.0 | Vastined | 0 | 2.2k |
| yujiepan/step3-vl-tiny-random | finetune | — | 3.16M | S | — | yujiepan | 0 | 755 |
| cyankiwi/Step3-VL-10B-AWQ-8bit | quantized | AWQ | 5.18B | M | apache-2.0 | cyankiwi | 1 | 554 |
| Iori2333/Step3-VL-10B-FP8 | quantized | -FP8 | 10.2B | L | apache-2.0 | Iori2333 | 1 | 458 |
| tiny-random/step3-vl | finetune | — | 3.16M | S | — | tiny-random | 0 | 170 |
| RimeRS/Huihui-Step3-VL-10B-abliterated | finetune | — | 10.2B | L | apache-2.0 | RimeRS | 0 | 71 |
| FenomAI/Step3-VL-10B-AWQ-4bit | quantized | AWQ | 4.31B | M | apache-2.0 | FenomAI | 0 | 51 |
| Date | Commit | SHA |
|---|---|---|
| 2026-02-04 | FP8 Quantized support (#21) | 5026053 |
| 2026-01-29 | Update modeling_step_vl.py | 821317c |
| 2026-01-23 | add sglang support (#18) | 4bdf973 |
| 2026-01-22 | Update README.md | 1028f3d |
| 2026-01-20 | add vllm image | 6fa640f |
| 2026-01-20 | add generation config and update Readme | 303e8de |
| 2026-01-19 | update notice for inference fix | 825bb1a |
| 2026-01-19 | update eos token | abdf361 |
| 2026-01-19 | update import modules | a77f584 |
| 2026-01-19 | modify import modules | 1ffe28d |
| 2026-01-19 | fix extra param | 99836ad |
| 2026-01-17 | Update README.md | 642dcd0 |
| 2026-01-16 | remove some unuse code | f7bf7c1 |
| 2026-01-15 | Update README.md | 82465dc |
| 2026-01-15 | Update README.md | 0c70a75 |
| 2026-01-15 | Update README.md | 8d1fd44 |
| 2026-01-15 | Update README.md | e6715b3 |
| 2026-01-15 | Update README.md | 588f66d |
| 2026-01-15 | Update README.md | 54a1061 |
| 2026-01-14 | Upload performance.png | 33ed491 |
| 2026-01-14 | Update modeling_step_vl.py | 23ce87d |
| 2026-01-14 | Update README.md | 2ddcc49 |
| 2026-01-14 | update config.json | fc4a171 |
| 2026-01-14 | add model deploy | 049cfe1 |
| 2026-01-13 | Upload 2 files | ac3d9d9 |
| 2026-01-13 | Update README.md | c02d396 |
| 2026-01-13 | Upload folder using huggingface_hub | cda41e5 |
| 2026-01-13 | initial commit | 6e2dd79 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:41 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:13 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
