Compare LLM Models
Select 2-5 models to compare side by side.
Select models to compare
0 / 5 selected
Top downloads
Select 2-5 models
| Attribute |
stepfun
|
|---|---|
| Company | stepfun |
| Release Date | 2026-09-20 |
| Parameters (B) | 600 |
| Architecture | Sparse Mixture-of-Experts (MoE) |
| Context Window | 1000000 |
| Input Modalities | text, image, video |
| Open Source | No |
| License | |
| Score | 44 |
| VRAM (GB) | 1200 |
| Compute | 8× NVIDIA H100 80GB (tensor parallel, BF16) |
| Benchmark: Agents-Last-Exam | 29.5 |
| Benchmark: AutomationBench | 51 |
| Benchmark: BrowseComp | 88.7 |
| Benchmark: CyberGym | 84.7 |
| Benchmark: DeepSWE | 67.7 |
| Benchmark: GDPval-AA | 1571 |
| Benchmark: GPQA-Diamond | 93.5 |
| Benchmark: HLE | 59.4 |
| Benchmark: MMLU-Pro | 76 |
| Benchmark: MMMU-Pro | 76 |
| Benchmark: Terminal-Bench-2.1 | 85 |
| Benchmark: Toolathlon-Verified | 74.1 |
| Pricing (Input) | 1.00 |
| Pricing (Output) | 2.70 |