Step3 VL 10B

Step3 VL 10B

StepFun · Released Jan 20, 2026 Professional
Intelligence #9 / 571
82.0 Our Score
AA Index #221 / 375
15.5 Artificial Analysis
Input
— Not priced
Output
— Not priced
Context
— Not reported

Analysis Summary

Step3 VL 10B is StepFun's 10-billion-parameter vision-language model, offering multimodal input capability alongside text generation. Its GPQA score of 0.69 is high relative to its size class, and the ifbench score of 0.502 indicates moderate instruction following for structured content tasks. The intelligence index of 15.5 places it in the lower mid-tier.

For businesses, the vision capability is the primary differentiator, enabling image-grounded content generation, document parsing with visual elements, and lightweight multimodal workflows. However, coding performance is limited (index 13.9), agentic capability is weak (10.7), and no pricing data is available, making cost-benefit assessment difficult.

This model is best suited to teams needing a small, multimodal assistant for content tasks where image understanding adds value. It is not a fit for coding, agent orchestration, or complex reasoning workloads.

Assessed June 6, 2026

Editorial notes

Step3 VL 10B from StepFun is a compact vision-language model with a strong GPQA of 0.69 and reasonable instruction following, suited to lightweight multimodal content tasks rather than complex reasoning or coding.

Rankings consider pricing, capabilities, benchmarks, and real-world applicability and are refreshed as new models launch. Feedback?

Performance Profile

Intelligence2.8Technical1.6Value0Content5.6
Intelligence 2.8/10
Technical 1.6/10
Content 5.6/10
Value 0/10

How Step3 VL 10B compares

Step3 VL 10B ranks #221 of 375 AI models we track for overall intelligence, #193 of 312 for coding, #249 of 286 for agentic tasks. Step3 VL 10B is currently free to use via OpenRouter.

Performance Indices

Source: Artificial Analysis

15.5 Intelligence Index
13.9 Coding Index
10.7 Agentic Index

Benchmark Scores

Intelligence

GPQA Diamond 69% Graduate-level scientific reasoning
HLE 10.2% Humanity's Last Exam
SciCode 31.1% Scientific computing

Technical

TerminalBench Hard 5.3% Agentic terminal tasks
τ²-Bench 16.1% Conversational agent benchmark

Content

IFBench 50.2% Instruction following

Benchmark data from Artificial Analysis and Hugging Face

How does Step3 VL 10B stack up?

Compare side-by-side with other professional models.

Compare Models

Model Information

ProviderStepFun
Release Date January 20, 2026
Status Active

Leaderboard Categories

Frequently asked questions about Step3 VL 10B

How much does Step3 VL 10B cost?

Step3 VL 10B is currently available for free via OpenRouter.

Is Step3 VL 10B good for coding?

On our coding benchmark index, Step3 VL 10B ranks #193 of 312 models, placing it in the broader range of the field for code generation and debugging.

Who created Step3 VL 10B?

Step3 VL 10B is developed by StepFun and was released on January 20, 2026.