StepFun

Multimodal Step models

Tier

Moderate

Headquarters

Shanghai, China

Founded

2023

Flagship

Step

Models tracked

4

Artificial Analysis

Top intelligence

42.6

Artificial Analysis Index

Cheapest model

$0.00

blended /1M

Fastest output

196

tokens / sec

Flagship

StepFun’s top-scoring model on the Artificial Analysis Intelligence Index.

Most intelligent model

Step 3.7 Flash

Released 2026-05-29

Intelligence

42.6

Coding

37.1

Math

In $/1M

$0.20

Out $/1M

$1.15

Speed t/s

175

Benchmark breakdown

GPQA Diamond
81
Humanity’s Last Exam
20
SciCode
40
τ²-Bench (agentic)
99
Terminal-Bench Hard
36
IFBench
67

Models & benchmarks

Every StepFun model on file — intelligence, coding, price, and speed.

ModelReleasedIntelCodingMathIn $/1MOut $/1MSpeed t/sLatency s
Step 3.7 Flash2026-05-2942.637.1$0.20$1.151750.97
Step 3.5 Flash 26032026-04-0238.534.6$0.10$0.301961.31
Step 3.5 Flash2026-02-0237.831.6$0.10$0.301781.03
Step3 VL 10B2026-01-2015.513.9$0.00$0.0000.00
Via Artificial Analysis

Trajectory

StepFun’s model releases over time, by intelligence.

Jan 2026 · 15.5▲ +27.1 indexMay 2026 · 42.6

House strengths

Where StepFun’s lineup over-indexes, averaged across its models.

Overview

A Chinese foundation-model startup building the multimodal Step family across text, vision, and audio.

Visit StepFun