StepFun: Step3

Image input Text input Text output
Author's Description

Step3 is a cutting-edge multimodal reasoning model—built on a Mixture-of-Experts architecture with 321B total parameters and 38B active. It is designed end-to-end to minimize decoding costs while delivering top-tier performance in vision–language reasoning. Through the co-design of Multi-Matrix Factorization Attention (MFA) and Attention-FFN Disaggregation (AFD), Step3 maintains exceptional efficiency across both flagship and low-end accelerators.

Key Specifications
Cost
$$$$$
Context
65K
Parameters
321B (Rumoured)
Released
Aug 28, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Top P Frequency Penalty Tool Choice Reasoning Tools Include Reasoning Temperature
Features

This model supports the following features:

Tools Reasoning
Performance Summary

StepFun: Step3, a cutting-edge multimodal reasoning model, demonstrates a unique performance profile. Despite its advanced architecture, including a Mixture-of-Experts design with 321B total parameters and 38B active, it tends to have longer response times, ranking in the 3rd percentile for speed across benchmarks. Similarly, its pricing is positioned at premium levels, placing it in the 20th percentile for cost competitiveness. However, Step3 excels in reliability, boasting an exceptional 99% success rate, indicating minimal technical failures and consistent provision of usable responses. In terms of specific benchmarks, Step3 achieved 99.0% accuracy in Ethics (53rd percentile), though at a high cost and very long duration. Its strongest performance was in Reasoning, where it scored 86.0% accuracy (84th percentile), showcasing strong complex problem-solving capabilities, albeit with the longest duration among all tests. In Email Classification, it achieved 96.0% accuracy (40th percentile) with a more moderate cost and duration. Overall, Step3's key strengths lie in its high reliability and strong reasoning abilities, while its primary weaknesses are its slow processing speed and premium pricing.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.57
Completion $1.42

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
SiliconFlow
SiliconFlow | stepfun-ai/step3 65K $0.57 / 1M tokens $1.42 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Free Executions Accuracy Cost Duration