Step3 VL 10B
StepFunOpen WeightApache 2.0 · Commercial OK
Description
STEP3-VL-10B is a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. Built on a unified, fully unfrozen pre-training strategy on 1.2T multimodal tokens integrating a language-aligned Perception Encoder with a Qwen3-8B decoder. Features Parallel Coordinated Reasoning (PaCoRe) to scale test-time compute for complex perceptual reasoning.
Release Date
2026-01-20
Parameters
10.0B
Context Length
—
Modalities
—
Capability Radar
10
general
31
coding
69
reasoning
44
science
58
agents
85
multimodal
Rankings
| Domain | #Rank | Score | Source |
|---|---|---|---|
| Code Ranking | 544 | 4.0 | AA |
| General Ranking | 477 | 24.0 | AA |
| Science | 272 | 47.0 | AA |
Benchmark Scores (LLM Stats)
(LLM Stats (zeroeval))Communication
Multi-Challenge
62.6%SR
General
MMMU
78.1%SR
Math
AIME 2025
87.7%SR
MathVista
84.0%SR
MathVision
70.8%SR
Multimodal
MMBench
91.8%SR
AA Evaluation Indices
(Artificial Analysis)Intelligence Index(Artificial Analysis)9.3
Gpqa(NYU + Cohere + Anthropic (2023))0.7
Ifbench(Google Research (2023))0.5
Scicode(UIUC + Argonne National Lab (2024))0.3
Tau2(Sierra + U Toronto + Vector Institute (2025))0.2
Hle(Center for AI Safety + Scale AI (2025))0.1
Terminalbench Hard(Stanford × Laude Institute (2026))0.1
Lcr(Artificial Analysis)0.0
LLM Stats Category Scores
(LLM Stats (zeroeval))Math80
Multimodal80
Reasoning80
General80
Healthcare80
Vision80
Communication60
Pricing
Input PriceFree
Output PriceFree
Blended Price (3:1)Free
Speed
Tokens/sec0.0
Time to First Token0.00s
Time to Answer0.00s
Provider Price Ranking
No provider data available