I got early access to StepFun's flagship model Step 5 Preview
This is a AI post classified by Jev as Frontier models (a model release), kept by the AI Radar because it carries real work, not commentary.
I got early access to StepFun's flagship model Step 5 Preview - Sparse Mixture-of-Experts (MoE) , 600B total parameters, but only 27B active per token. - 1M tokens with native vision input. I tested it inside an autonomous agent harness to see how it handles end-to-end coding. I gave it a detailed PRD to build a Creator CRM with database from scratch. at least on the agentic coding side, my personal feeling is that it feels very close to GLM-5.3 and Kimi K3 level model in terms of reasoning depth and staying on track. I haven't tested it across other domains yet, but for long sessions it hol
Posted by AshutoshShrivastava (81.5k followers) 1 h ago · 27 likes · 2.8k views · view the original post on X. Kept by the AI Radar as Frontier models.
More AI work like this
- @AdamHoltererer — @AdamHoltererer
- GPT-6 Astra Pro makes a DevDay hype reel — @AdamHoltererer
- Effective Altruists seeing this: — @Hesamation
- 🚨 Grok 4.7 looks to have been called through the xAI API via Hermes — @LuminaBench
- Grok 4.7 Leak: Spotted In A Cursor Agent PR 👀 — @Priyannkaaaa
- 哇，看起来 GPT 6 Luna 真的要来了 — @Fei2411
- 🚨 GPT-6 Luna appears to be already showing up in Codex backend responses — @LuminaBench
- New model: qwen3.8-omni-flash-realtime — @aitrackerbot
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 38.8k posts from 5k X accounts over the last 14 days, 1.6k tools, 19 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-21 15:24 UTC. Full method.