Step 5 Preview

Name: Step 5 Preview model facts
Creator: OSMEXO

stepfun/step-5-preview

StepFun
Family · Step 5
Released 18 Sep 2026
Open weights
Text · Image · Video → Text
Reasoning · adjustable
Tools · JSON · Vision

Not on OSMEXO yet

Request this model

Spec & list price

Context1M

Max output64K

Input / 1M$1.00

Output / 1M$2.702.7× input

Scores · standing

Intelligence43.7#24/489

LiveBench58.6#24/32

Usage

Tokens / day799.7B▲ 63% 7d

Usage rank#8

Requests / day1.7M

Tool calls / req0.45

Per request · 30 days

Avg input208.9K14× median

Avg output1.1K1.6× median

Cache hit95%1.4× median

Reasoning share22%0.4× median

Tool-call failures1.7%0.9× median

Consumption

Peak 3.7T 14 OctLow 4.4B 17 OctRank today #8

News & trends

4 mentions this week ▼ 80%this model · videos, posts, articles · 30 days

Trending · 14 days

Benchmarks

bar = share of the top score · Benchmark data: Artificial Analysis

Long-context reasoning#2 of 393

88.3%

SciCode#15 of 160

58.9%

Humanity's Last Exam#23 of 465

46.5%

Intelligence index#24 of 489

43.7

Terminal-Bench 4.0#21 of 152

33.3%

Speed · median across 3 providers, latest day

Output speed#63/334105 tok/s1.6× faster than median

Fastest provider129 tok/sbest of 3 providers

First token#2511.85 s1.8× longer than median

First answer token#5615 sthinking included

#8 cheapest of 15 with a similar score

Value

Intelligence Index

Blended price per 1M tokens (log)

LiveBench suites

Browser24.8

Coding85.6

Data Analysis63.5

Design76.9

Long Agentic19.4

Reasoning83.8

Research72.8

▲ Strongest: Coding, Reasoning. ▼ Weakest: Long Agentic (19).

Technical details

Reasoning: on · medium
Tokenizer: Other
License: other
Parameters: 604B
Moderation: none
Downloads · likes: 5.2K · 137

Reasoning effort

high
medium · default
low

Request parameters 19

tools
tool_choice
response_format
structured_outputs
reasoning
reasoning_effort
include_reasoning
temperature
top_p
top_k

+9 more

min_p
max_tokens
stop
seed
frequency_penalty
logprobs
presence_penalty
repetition_penalty
top_logprobs

Interesting facts

Less output room than similar models (64.0K against a 129.5K median).
Released 45 days ago.

When to use

Best for

Very long document reasoning
Coding and scientific problem solving
Image and video understanding
Hard analysis with adjustable reasoning effort

Think twice for

Browser-driving agents
Multi-hour autonomous agent runs
Production use needing a stable release
High-volume simple chat

Alternatives

other vendors · similar capability

Gemini 3.8 Flash
Google
40.9
MiMo-V2.6-Pro
Xiaomi
46.3
GPT-6 Sol
OpenAI
47.6
Grok 4.7
xAI
46.4

Want Step 5 Preview on OSMEXO?

Request this model on OSMEXO