StepFun Launches Step 5 Preview: 600B MoE Flagship, Open-Weights in October

StepFun officially released its next-generation flagship foundation model, Step 5 Preview, on September 20, positioned for agentic work and professional knowledge work. Built on a sparse MoE architecture with 600B total parameters and 27B activated per inference, it natively supports text and visual input, offers a 1M-token context window, and will have its weights open-sourced on October 15. The model had previously been quietly benchmarked by Artificial Analysis, sparking discussion, and the official release confirmed most of the leaked specs.

Confirmed

• Architecture and specs: 600B total / 27B activated MoE, 1M context, multimodal (native visual input) (m6, m7, m9, m10, m11)
• Benchmark results: scored 44 on the Artificial Analysis Intelligence Index v4.2, on par with Kimi K3 (max), placing it in the global top three among open-source models and on the cost/intelligence Pareto frontier (m1, m9, m10)
• Pricing: ¥7 input / ¥20 output / ¥0.35 cache (per million tokens), roughly one-third to one-fifth of Kimi K3's price (figures vary slightly across posts) (m11, m1, m7)
• Open-source timeline: weights to be released October 15 (m6, m11)
• Capabilities: officials say it can execute long-horizon tasks across software engineering, web apps, 3D workflows, and programmable hardware environments, running continuously for up to 24 hours; tasks include GPU kernel optimization, with the ability to remember previous results and iterate based on execution feedback (m2)
• Finance: features its own FinStepBench benchmark covering LiveSearch, CorporateValuation, and DeepResearch task types; in one case, a single agent action coordinated 950 web scrapes, consolidated roughly 300,000 monthly records over 25 years across 1,000 locations, and produced a 17-table analytical workbook with source reconciliation, formulas, and trend models (m4, m5)
• Ecosystem: open trial and docs, a Discord community, and a platform offering a limited-time free Step subscription plus the Step audio realtime 3 voice model, with one API key covering both language and voice (m8)

Why it matters

Step 5 Preview matches Kimi K3's intelligence score at about one-third of the price, pushing open-source models to a new capability-cost Pareto frontier. Combined with weights opening on October 15, it will directly shake up the domestic and international open-source model landscape. Its focus on 24-hour long-horizon agentic tasks and finance-specific benchmarks also signals Chinese models going deeper from general chat toward professional workflows.

2026-09-19 ~ 2026-09-20 · 13 related posts

• StepFun's Step 5 Preview matches Kimi K3 at one-third the price, per Artificial Analysis — No-Head-Royal · 2026-09-19
• Step 5 Preview Quietly Debuts on AA: Score 44, 1M Context, $1/1M Input — teortaxesTex · 2026-09-19
• StepFun's Step 5 Preview Surfaces: 600B Params, Output Priced at 1/5 of Kimi K3 — teortaxesTex · 2026-09-20
• [source] Step 5 Preview Sustains 24-Hour Agent Runs, From GPU Kernels to Post-Training — StepFun_ai · 2026-09-20
• [source] StepFun Launches Step 5 Preview: 600B/27B MoE Flagship, Open Weights Oct 15 — StepFun_ai · 2026-09-20
• Step 5 Preview: 950 Web Fetches and 17-Sheet Workbooks in One Agent Action — StepFun_ai · 2026-09-20
• StepFun's FinStepBench Puts Step 5 Preview Through Financial Scrutiny — StepFun_ai · 2026-09-20
• [source] Step 5 Preview Now Open to Try, Step Plan Subscription Free for Now — StepFun_ai · 2026-09-20
• StepFun's Step5Preview: 600B-param MoE with 27B active, top-3 open-source on AA Index — APPSO · 2026-09-20
• StepFun releases Step5 Preview: 600B/27B MoE, 1M context, beats Opus 5 on kernel optimization — 赛博禅心 · 2026-09-20
• StepFun's Step5 Preview scores 44 on AA Index, ranking top-3 among open models — 新智元 · 2026-09-20
• StepFun Launches Step 5 Preview: 600B MoE with 27B Active, Open Weights on Oct 15 — airesearch12 · 2026-09-20
• StepFun's Step-5 Preview Leaks: 600B/27B Active, 1M Context, GPQA Diamond 93.5 (Vendor Numbers) — airesearch12 · 2026-09-20