• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI
Report

Open weights for StepFun’s 600B model are reportedly planned for October 15

A user describes a mixture-of-experts model with 27 billion active parameters, vision and a 1M context window, designed for coding, 3D and web design.

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)T(
37 Sources, 22d ago, first seen 22d ago

TLDR

A user writing on September 21 describes StepFun’s model as having 600 billion total parameters, with 27 billion active, and says open weights are planned for October 15, 2026. The user says ArtificialAnalysis puts it on par with Kimi K3 in intelligence. They also describe it as designed to be relatively inexpensive compared with its direct competitors—not as an established price advantage.

Combined views

146

37 Sources, first seen 22d ago

2 reposts

Combined views

146

37 Sources, first seen 22d ago

2 reposts

Sentiment

Positive60.4%39.6%Negative

Summary

Positive accounts welcomed AI agent features like Muse’s Shop Pay integration and StepFun’s preview for their practical coding and shopping benefits, while negative replies questioned AGI claims and the nonstop pace of new model releases.

Based on 43 sentiment-bearing replies from 41 accounts across 12 conversations.

Sentiment

Positive60.4%39.6%Negative

Summary

Positive accounts welcomed AI agent features like Muse’s Shop Pay integration and StepFun’s preview for their practical coding and shopping benefits, while negative replies questioned AGI claims and the nonstop pace of new model releases.

Based on 43 sentiment-bearing replies from 41 accounts across 12 conversations.

37 Sources

Fanfulla@Fanfulladev@StepFun_ai just dropped the Step 5 Preview, its new flagship model for agentic work. It has: - 600B total / 27B active MoE - 1M context + Vision - Built for long-horizon agentic execution - Focus on software engineering, research and finance - Open weights coming October 15 - Same scoring on https://artificialanalysis.ai/ of Grok 4.6 and Kimi K3 Max I think the 27B active out of 600B is probably the most interesting part here and would love to see integrated on DS4 by @antirez After this i see Chinese labs don’t seem too worried about “pacing the frontier” right now22d
Muse Cases@MusecasesMeta is giving away 100 million free tokens a week and billion-token invite codes. Not generosity, just the most expensive customer acquisition in tech history. Full breakdown inside. https://x.com/i/article/210180022573190348820d
Nikhil Pathak@web3devopUpcoming AI models : - Grok 4.7 - Gemini 4 - Claude Opus 5.5 Which one do you try first ?20d
Mia@MiaAI_labStepFun is so back! I loved StepFun 3.7 Flash back when it was released, I'm so glad they're returning to the game. Seems like a very strong model! - 600B total and 27B active MoE - 1M context with Vision - Designed to perform extremely well on coding, 3D and web design. ArtificialAnalysis puts it on par with Kimi K3 in Intelligence, while Kimi has over 4x as many params! It's also designed to be relatively inexpensive compared with the models it's directly competing against. Open weights on October 15th. Can't wait to try it out!20d
HypeR🎖️@ProsHyperAI in the last 24 hours: • Grok 4.7 ships for coding • StepFun drops a 600B agent model • SoftBank raises ~$11B for OpenAI • UN panel: don’t wait to regulate agents • US–China talk AI incident alerts20d
Nifal Adam@nifal_adam🧵2/14 @stepfun_ai StepFun has launched Step 5 Preview, a 600B-total-parameter agent model. StepFun says the model targets software engineering, professional knowledge work, and finance. Open weights are planned for October 15, 2026. https://www.gobignews.com/ai/stepfun-launches-step-5-preview-a-600b-total-parameter-agent-model20d
Agentic Core AI | AI News@AgenticCoreAI🌙 AI ALPHA NIGHT | September 21, 2026 SoftBank drops $11B bonds for OpenAI. US-China lock in AI incident talks. Altman heads to the UN. Gates pushes language equity. New hardware and standards moves close the day. 👇20d
Pradeep Kapoor@pradeepXkapoorAGI IS HERE!! OpenAI dropped one of the craziest AI updates of 2026. They say a new internal model that started training on August 28 has already: - produced the Navier-Stokes Millennium Prize result - resolved 100+ long-standing open problems - made progress across most areas of mathematics And apparently the pace surprised even OpenAI’s own mathematicians. They’ve now helped set up an independent group of top mathematicians just to figure out how these results should be reviewed and released. AI doing actual research is now absolutely real20d
Xian Zeng@zeng_xian40157Today in AI: 1. OpenAI model solves 100+ open math problems 2. Grok 4.7: 2x faster, 50% cheaper 3. Treasury: HF breach on OpenAI mgmt 4. Meta's Muse overtakes ChatGPT on App Store Capability is racing; guardrails are finally written in public. #AI #MachineLearning #Tech20d
🚨 AI News | TestingCatalog@testingcatalogMeta ❤️ Shopify Muse agent can now use Shop Pay on Shopify stores, enabling agentic shopping in countries where Muse is supported. It is hard to imagine how huge the agentic shopping market will be in a few years, given that small businesses on Facebook or Instagram will soon be able to integrate out of the box. That's a whole new world 🤖20d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    37 Sources

    Fanfulla@Fanfulladev@StepFun_ai just dropped the Step 5 Preview, its new flagship model for agentic work. It has: - 600B total / 27B active MoE - 1M context + Vision - Built for long-horizon agentic execution - Focus on software engineering, research and finance - Open weights coming October 15 - Same scoring on https://artificialanalysis.ai/ of Grok 4.6 and Kimi K3 Max I think the 27B active out of 600B is probably the most interesting part here and would love to see integrated on DS4 by @antirez After this i see Chinese labs don’t seem too worried about “pacing the frontier” right now22d
    Muse Cases@MusecasesMeta is giving away 100 million free tokens a week and billion-token invite codes. Not generosity, just the most expensive customer acquisition in tech history. Full breakdown inside. https://x.com/i/article/210180022573190348820d
    Nikhil Pathak@web3devopUpcoming AI models : - Grok 4.7 - Gemini 4 - Claude Opus 5.5 Which one do you try first ?20d
    Mia@MiaAI_labStepFun is so back! I loved StepFun 3.7 Flash back when it was released, I'm so glad they're returning to the game. Seems like a very strong model! - 600B total and 27B active MoE - 1M context with Vision - Designed to perform extremely well on coding, 3D and web design. ArtificialAnalysis puts it on par with Kimi K3 in Intelligence, while Kimi has over 4x as many params! It's also designed to be relatively inexpensive compared with the models it's directly competing against. Open weights on October 15th. Can't wait to try it out!20d
    HypeR🎖️@ProsHyperAI in the last 24 hours: • Grok 4.7 ships for coding • StepFun drops a 600B agent model • SoftBank raises ~$11B for OpenAI • UN panel: don’t wait to regulate agents • US–China talk AI incident alerts20d
    Nifal Adam@nifal_adam🧵2/14 @stepfun_ai StepFun has launched Step 5 Preview, a 600B-total-parameter agent model. StepFun says the model targets software engineering, professional knowledge work, and finance. Open weights are planned for October 15, 2026. https://www.gobignews.com/ai/stepfun-launches-step-5-preview-a-600b-total-parameter-agent-model20d
    Agentic Core AI | AI News@AgenticCoreAI🌙 AI ALPHA NIGHT | September 21, 2026 SoftBank drops $11B bonds for OpenAI. US-China lock in AI incident talks. Altman heads to the UN. Gates pushes language equity. New hardware and standards moves close the day. 👇20d
    Pradeep Kapoor@pradeepXkapoorAGI IS HERE!! OpenAI dropped one of the craziest AI updates of 2026. They say a new internal model that started training on August 28 has already: - produced the Navier-Stokes Millennium Prize result - resolved 100+ long-standing open problems - made progress across most areas of mathematics And apparently the pace surprised even OpenAI’s own mathematicians. They’ve now helped set up an independent group of top mathematicians just to figure out how these results should be reviewed and released. AI doing actual research is now absolutely real20d
    Xian Zeng@zeng_xian40157Today in AI: 1. OpenAI model solves 100+ open math problems 2. Grok 4.7: 2x faster, 50% cheaper 3. Treasury: HF breach on OpenAI mgmt 4. Meta's Muse overtakes ChatGPT on App Store Capability is racing; guardrails are finally written in public. #AI #MachineLearning #Tech20d
    🚨 AI News | TestingCatalog@testingcatalogMeta ❤️ Shopify Muse agent can now use Shop Pay on Shopify stores, enabling agentic shopping in countries where Muse is supported. It is hard to imagine how huge the agentic shopping market will be in a few years, given that small businesses on Facebook or Instagram will soon be able to integrate out of the box. That's a whole new world 🤖20d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet