Leaderboard / StepFun
Step3-VL-10B
StepFun · Step3-VL · released 2026-01 · data checked 2026-11-01 Needs verification
Step3-VL-10B is a 10B-parameter dense model from StepFun (China), released 2026-01 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 8 GB of memory at 4-bit with an 8K-token context.
StepFun's only open model under 100B; Step 3.5/3.7 Flash and Step 5 are 196B+.
Benchmarks
Each bar shows this model's score. The tick marks the best small model on that benchmark.
No benchmark results for this release yet. Know of one? Send it.
Specifications
- Lab
- StepFun (CN · China)
- Architecture
- Dense transformer
- Total parameters
- 10B
- Active parameters
- All (dense)
- Size class
- Small
- Context window
- 64K tokens
- Input modalities
- text, image
- Reasoning
- Reasoning (thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~8 GB at 4-bit, 8K context (more sizes)
Compare with
- Step3-VL-10B vs MiniCPM-SALA (9.5B, index 52.6)
- Step3-VL-10B vs Qwen3.5 9B (9B, index 63.9)
- Step3-VL-10B vs Nemotron Nano 9B v2 (8.9B, index 52.5)
Sources
Something wrong here? Report it.
Run it locally
No quantized build from a well-known publisher yet; check the model card for instructions.
Memory needed: weights, context cache and runtime buffers (context cache estimated: attention layout not published).
| Quantization | 4K context | 32K context | 64K context |
|---|---|---|---|
| 16-bit F16 | 21 GB | 25 GB | 30 GB |
| 8-bit Q8_0 | 12 GB | 16 GB | 21 GB |
| 4-bit Q4_K_M | 7.4 GB | 12 GB | 16 GB |