Skip to main content
Back to News
Technology
2 min read
CN

StepFun's Step 5 Preview Climbs to Global No. 2 Open-Source Model, Passing Kimi K3

The AMW Read

A one-day leaderboard reshuffle extending yesterday's top-3 debut for an already-tracked player, notable mainly for sharpening the intelligence-per-active-parameter tradeoff among competing Chinese open-weight labs.
NoveltySignificance
Foundation Models · Player MapScaling Laws
Stepfun
Stepfun

Foundation Models / LLMs

View Company Profile

StepFun's Step 5 Preview Climbs to Global No. 2 Open-Source Model, Passing Kimi K3

StepFun's newest flagship model, Step 5 Preview, moved from third to second place on Artificial Analysis' global open-source Intelligence Index in a single day, per the leaderboard's September 20 update (version 4.3.2), overtaking Moonshot AI's Kimi K3. The 600-billion-parameter mixture-of-experts model activates only 27 billion parameters per task, a smaller footprint than rival models scoring at similar intelligence levels. Artificial Analysis said it ran more than ten repeated tests per benchmark to verify the result. Step 5 Preview is already live, positioned for AI coding, software engineering, professional knowledge work, and finance use cases, with open-weight release planned for October 15.

The move extends a story we flagged just a day earlier, when Step 5 Preview first broke into the open-source top three at roughly one-eighth Claude Opus 5's per-task cost. Passing Kimi K3 puts StepFun ahead of another well-capitalized Chinese lab on a leaderboard that has reshuffled twice inside 48 hours — the competitive gap among Alibaba, Zhipu AI, Moonshot AI, and StepFun is now measured in days. What's moving is less raw capability than the ratio of intelligence to active-parameter cost, which sets how cheaply a model can serve the request volumes agentic applications generate.

For builders evaluating open-weight models for coding or finance agents, Step 5 Preview's lower active-parameter count is the figure to watch once weights publish October 15 — it shapes inference cost at scale, not just leaderboard position. Per the AI Market Watch index, StepFun has raised a cumulative $3.8 billion since its 2023 founding (index coverage, not a census), giving it runway to keep contesting a leaderboard that is turning over within days rather than settling.

#StepFun #Step5Preview #OpenSourceAI #ArtificialAnalysis #ChineseAI #LLM

#StepFun#Step 5 Preview#Artificial Analysis#open-source leaderboard#Kimi K3

How This Connects

Based on Foundation Models · Player Map

  1. 1d agoThinking Machines releases Inkling 2 Mini for local use on laptopsThinking Machines
  2. 1d agoOpenAI reports chip design gains on Jalapeno and AMD hardwareOpenAI
  3. 6d agoTranslated releases Lara Mini Base, a multilingual 7.2B modelTranslated
  4. 1w agoDeepSeek Reportedly Raises $15 Billion With Tencent and CATL ParticipationDeepSeek
  5. 1mo agoReka releases Flint for local coding, citing AI-led model developmentReka
  6. 2mo agoStepFun's Step 5 Preview Climbs to Global No. 2 Open-Source Model, Passing Kimi K3 · THIS ARTICLE

Related News

More news from Stepfun

Stay updated with the latest news and announcements from Stepfun.

View all Stepfun news

Discover AI Startups

Explore 5,000+ AI companies with VC-grade analysis, funding data, and investment insights.

Explore Dashboard