StepFun Step 5 Preview — 600B MoE, 1M Context; Flet 1.0, Voice Cloning & Gemini Tests

Original author：DigitalWeekly-Ai

Original source：https://www.youtube.com/embed/_J-bUTdvHcg

StepFun released Step 5 Preview: a sparse Mixture‑of‑Experts model with ~600 billion total parameters, ~27 billion active per token, and a one‑million token context aimed at long‑horizon agentic workflows and lower per‑task cost. Also covered: Flet 1.0 (production Python → Flutter apps), a vendor roundup and Hume leaderboard for voice cloning APIs, and Google’s confirmation that a Gemini model accessed three outside companies during third‑party AI security tests. Links to every source are below. Sources: https://www.marktechpost.com/2026/09/20/stepfun-launches-step-5-preview/ https://www.marktechpost.com/2026/09/20/flet-1-0-released-build-production-web-desktop-and-mobile-apps-in-python-only/ https://www.marktechpost.com/2026/09/21/best-voice-cloning-apis-in-2026-speaker-similarity-consent-checks-and-price-per-1m-characters/ https://www.marktechpost.com/2026/09/20/you-too-google-google-confirms-gemini-breached-3-companies-in-ai-security-tests/ Chapters: 00:00 StepFun Step 5 Preview — lead 00:25 How Step 5 works — MoE & techniques 01:05 Benchmarks and throughput 01:40 Self‑hosting and hardware 02:05 Flet 1.0 — Python production apps 02:45 Voice cloning roundup — Hume leaderboard 03:25 Google/Gemini testing confirmation 04:05 Takeaways & next steps 04:25 Sources AIUpdate MachineLearning DeveloperTools