System Card
← leaderboard
Stepfun (阶跃星辰)
StepFun
Step
#18
of 28 · general
Step 5 Preview
600B
params · MoE
1M
context
$1
/
$2.7
per 1M in / out
2026-09-18
released
50
100
Coding 25 · #22 of 28 · 26% of the recipe
Agents 32 · #22 of 29 · 47% of the recipe
Cyber: no reading
Engineering 23 · #17 of 26
Math 19 · #23 of 26 · 33% of the recipe
Health 13 · #21 of 24 · 38% of the recipe
Legal 35 · #19 of 27
Professional 26 · #22 of 28
Writing 14 · #16 of 21 · one source
Instruct 21 · #17 of 25 · 41% of the recipe
Hallucination 47 · #10 of 26
Multimodal 17 · #19 of 24 · 29% of the recipe
Coding
25
#22
Agents
32
#22
Cyber
—
Engineering
23
#17
Math
19
#23
Health
13
#21
Legal
35
#19
Professional
26
#22
Writing
14
#16
Instruct
21
#17
Hallucination
47
#10
Multimodal
17
#19
DeepSeek Pro
MiniMax Flash
GLM Flash
MiMo Flash
GPT Luna
Measured
≈ imputed · amber dot = lab-self-reported among the sources
HLE
46.5%
#15
SciCode
58.9%
#10
CritPt
20.9%
#12
APEX Agents
38.0%
#18
GDPval
54.2%
#16
Arena
1447
#16
AA Index
43.7
#14
Vals Index
49.4
#19
Neighbors
DeepSeek Pro
→ stronger math · pricier
MiniMax
MiniMax Flash
→ stronger agent · narrower
Zhipu (智谱)
GLM Flash
→ narrower · cheaper
Xiaomi
MiMo Flash
→ less factual · cheaper
GPT Luna
→ stronger math · cheaper