Step-3
Released 2025-07 · hybrid · 66K tokens · 4 benchmarks · Open weight
Editorial notes
Multimodal MoE 321B / 38B activos. MMMU 74.2 lidera open-source VLM.
Spec sheet
- Company
- StepFun
- Country
- CN
- Type
- hybrid
- Release
- 2025-07
- Context
- 66K tokens
- Params total
- 321B
- Params active (MoE)
- 38B
- License
- apache-2.0
- Quants
- BF16, Q8_0, Q5_K_M, Q4_K_M
- Slug
- step-3
Benchmarks (4)
Reasoning 2
Cite this model
BibTeX · APA
BibTeX
@misc{frontier-step-3,
title = {Step-3},
author = {{StepFun}},
year = {2025},
note = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
url = {https://frontierbenchmarks.com/models/step-3}
} APA
StepFun (2025). Step-3 [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/step-3
Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.