Compare AIFind AIAI NewsAI How-To
About Us
PrivacyTermsFAQContactContact
AIB Inc.Company info
© 2026 AIB Inc.
Stepfun
Stepfun

Step-5

Release Date 2026-09-20··1M INProprietary Model
Overall3.0Top 25%
AutomationResearching
Office4.0Top 9%Coding3.0Top 27%

Step-5 is Stepfun’s large flagship AI model, designed to work with both text and visual inputs across complex technical projects. Rather than focusing primarily on short conversations, it is built for practical, multi-step engineering and research tasks. For example, users can provide a photo of a bedroom and have the model create an interactive 3D room-planning tool, or assign it to run automated optimization processes on high-performance GPUs over the course of a full day. It can also support large-scale information gathering and organization, such as collecting data from hundreds of online sources to build climate datasets covering thousands of locations.

The model is primarily aimed at software developers, data analysts, and finance professionals working on tasks such as automated problem-solving, code refactoring, and verifiable financial valuation models. Stepfun moved directly from Step-3.7-Flash to Step-5 without releasing a separate Step 4 generation, placing greater emphasis on long-running autonomous execution and the coordinated use of multiple tools than on everyday chat interactions. For the most demanding long-duration tasks, however, frontier models such as Claude Opus 5 may still offer advantages in some areas.

  1. 01ImagesGenerated images
  2. 02CostCost · Usage · Speed
  3. 03User CompareUser Comparisons
  4. 04Expert EvalExpert Performance Evaluation
  5. 05OpinionPublic Opinion
  6. 06Related Info

API cost

$1IN$2.7OUT
Calculate Cost

Output Speed

Slowper 2,000 tokens
First Output24.2s
total46.5s
90tok/s
First Output 24.2stotal: 46.5s
SourceArtificial AnalysisLast updatedOct 7, 2026

Expert Performance Evaluation

Overall performance

MoreAll overall performance AI rankings
43.71454

About AA Intelligence IndexMore about AA Intelligence Index

Automation (agents)

MoreAll automation AI rankings
QCan it pick the right tools and use them properly?MCP-Atlas·Top 11%#4
85.6%Reasoning effort: HighSelf-reported by developer

About MCP-AtlasMore about MCP-Atlas

Documents and office work

MoreAll documents and office work AI rankings
88.3%146859.0%

About Long Context ReasoningMore about Long Context Reasoning

Research and fact-checking

MoreAll research and fact-checking AI rankings
88.7%Reasoning effort: HighSelf-reported by developer59.4%

About BrowseCompMore about BrowseComp

Coding

MoreAll coding AI rankings
33.3%151158.9%1490

About Terminal-Bench 4.0More about Terminal-Bench 4.0

Writing and chat

MoreAll writing and chat AI rankings
14361460142114451420

About Arena Writing, Literature & LanguageMore about Arena Writing, Literature & Language

Math and science

MoreAll math and science AI rankings
46.5%1469

About Humanity's Last ExamMore about Humanity's Last Exam

Image understanding

MoreAll image understanding and analysis AI rankings
127876.0%

About Arena VisionMore about Arena Vision

Specialized fields

14821472144784.7%

About Arena Hard PromptsMore about Arena Hard Prompts

See each evaluation’s detail page for its source and last update·Data sources & removal requests