Step-3.7-Flash

A multimodal MoE model focused on agent efficiency, coding, search, and multimodal workflows.

An open-weight mixture-of-experts vision-language model release for agentic coding, search, and multimodal workflows.

Pricing

Model profile · Current snapshot

Input / 1M

$0.20

Output / 1M

$1.15

Blended / 1M

$0.438

Output TPS

85.99

TTFT (s)

1.76

Context window

256,000 tokens

Arena ranking

Benchmarkable

Yes

Model level

release

Intelligence Index

19.5

Coding Index

39.6

GPQA

0.81

HLE

0.21

SciCode

0.44

IFBench

0.67

LCR

0.74

TerminalBench Hard

0.36

TAU2

0.99

0 linked stories

No linked stories yet.