The name HySparse2 really makes me think of Hy (Hunyuan). I even catch myself reading it as Hunyuan-Sparse XD Ever since DeepSeek brought YOCO back into the spotlight, Chinese model backbones seem to have entered another wave of architectural evolution. In a way, DeepSeek has always been the big brother we can all learn so much from. Say Saint Liang plz 🙏 Really excited to see what surprises V4.1 Pro has in store for us. Of course, we can't afford to fall behind either. Scale up!

Join us in building a safer, more reliable, and high-performance Agent Harness! 🚀 Be water my friend 🍵 Water18 is a great name btw. In Chinese culture, water symbolizes wealth, and “18” sounds like “yao fa,” meaning “going to prosper.”

The Rearrangement Inequality and Its Generalizations Starting from the rearrangement inequality, this article follows two lines of development — "matricization" and "multi-sequence extension" — to survey its principal generalizations.

Step Plan Mini and Plus are BACK, with limited availability for now. We’ll keep opening more spots and posting updates as we go. Have fun building with StepFun models 🫡

StepFun's Step 5 Preview scores 44 on the Artificial Analysis Intelligence Index, matching Kimi K3 (max) at ~2.8x lower cost per task, but trails peers on agentic evaluations Step 5 Preview is @StepFun_ai's new flagship model, with 600B total and 27B active parameters, succeeding Step 3.7 Flash (released May 2026). It scores 44 on the Intelligence Index, level with Kimi K3 (max) and just behind GLM-5.3 (max, 45) and Qwen3.8 Max (45) Key takeaways: ➤ Step 5 Preview costs ~2.8x less per Intelligence Index task than models at the same score. It costs ~$0.72 per task, against ~$2.00 for Kimi K3 (max) at the same score of 44 and ~$2.01 for GLM-5.3 (max) at 45. This is driven by pricing: at $1/$2.70 per 1M input/output tokens, it is priced below both on input and output. MiMo-V2.6-Pro is the one model that scores higher (46) at a lower cost per task ($0.13) ➤ Frontier reasoning is the standout strength, and where the jump from Step 3.7 Flash is largest. Step 5 Preview scores 46% on Humanity's Last Exam, in line with Kimi K3 (max, 47%), and 21% on CritPt, between Kimi K3 (23%) and GLM-5.3 (max, 19%). Both are up sharply from Step 3.7 Flash: +25 points on HLE and +19 points on CritPt ➤ Higher AA-Omniscience accuracy than GLM-5.3 at fewer parameters, but with more hallucination. At 600B total parameters, Step 5 Preview reaches 42% accuracy on AA-Omniscience, our benchmark measuring factual recall and hallucination, ahead of GLM-5.3 (max, 34%, 753B) and behind Kimi K3 (max, 48%, 2.8T). It attempts more questions than GLM-5.3 (68% vs 55%) and hallucinates more often when it does (43% vs 30%), landing at 16 on the AA-Omniscience Index, between GLM-5.3 (14) and Kimi K3 (20) ➤ Agentic evaluations are where Step 5 Preview lags peers at a similar Intelligence Index score. It scores 1,566 Elo on GDPval-AA, our primary evaluation for agentic performance, behind Qwen3.8 Max (1,668) and GLM-5.3 (max, 1,646). The gap holds on Terminal-Bench 4.0 (33% vs 39% and 42%), AA-Briefcase (1,432 Elo vs 1,640 and 1,525) and AutomationBench-AA (51% vs 56% and 62%) Key model details: ➤ Model Size: 600B total parameters, 27B active MoE model ➤ Context window: 1M tokens ➤ Multimodality: Text, image and video input, text output ➤ Pricing: $1/$2.70 per 1M input/output tokens, with cached input at $0.05/M ➤ Availability: StepFun first-party API, with open weights release planned for October 15th ➤ Licensing: Closed weights currently, with weights release planned for October 15th

Nah, we’re still hanging in there. Thanks for the load test tho🤣 For everyone who hasn’t been able to get a plan: we hear you 🥲 Step Plans are temporarily sold out, but the team is already working hard to bring them back. I’ll post here as soon as they’re available again — thanks for your patience!

Thank you all for the support since the Step 5 Preview launch. I honestly didn’t expect to see so many new people joined our Discord and start building with our new model this quickly. here’s a quick update regarding the question received: - Step Plan subscriptions are temporarily unavailable, and we will share an update soon on when they reopen. - The API experienced some instability yesterday. You should be able to use it. Our new model matters, but so does the developer experience. I will keep you guys posted.

Now you can create videos the way you vibe-code. Meet Pexo, your AI video agent. Share your vision and references. Work through the details with Pexo like you’re messaging a friend. No complex prompts. No new tools to learn. Want to change something? Mark it on the video and tell Pexo what to change, just like leaving a comment in a doc. Everything in this video was made with Pexo: the visuals, the motion graphics, the music, the captions, and the voiceover. Even our founder @evanLiaoQ appears on screen. You vibe-coded what you built. Now vibe-create how the world sees it. #Pexo# #AIVideoAgent# #VibeCreate# #MadewithPexo#

Thanks everyone for supporting @StepFun_ai. We’re urgently scaling up our online capacity and hope to get our latest model into more hands soon. We’re also working on a proposal to offer free Pro subscriptions to a number of PhD students and researchers once we have enough capacity. We’d love your help coming up with real-world problems that could push the model further, across fields like finance, healthcare, legal, engineering, and manufacturing. Drop your ideas in the comments or reach out directly. We’ll consider the number of slots based on the resources available. Toward AGI, guys! 🚀

🚨BREAKING: China's StepFun has officially launched Step 5 Preview, a 600B MoE model with just 27B active parameters and a 1M-token context window. Step 5 Preview matches Kimi K3 (max) at 44 on Artificial Analysis, but at about 65% lower cost per task. StepFun is now on the cost/intelligence Pareto frontier. The model is available via API now, and StepFun says it will release the weights on October 15. China keeps shipping

Step 5 Preview works across software environments and sustains execution over long horizons. Its capabilities extend from software engineering and web applications to 3D workflows and programmable hardware. Over longer horizons, Step 5 Preview keeps track of prior results, uses execution feedback to decide what to try next, and continues iterating. We test this behavior in runs lasting up to 24 hours, including tasks involving GPU kernel optimization and automated post-training.

Introducing Step 5 Preview: Advancing the Pareto Frontier. Step 5 Preview is our new flagship model for agentic work, delivering frontier-level performance across software engineering and professional knowledge work, with particular strength in finance. - 600B total / 27B active MoE, with 1M context + Vision - Substantially lower task cost at comparable intelligence - Broad software engineering capabilities with sustained execution over long horizons Try Step 5 Preview: Model page: Open weights on Oct 15.

We’ve been quiet for too long. It’s time for a change. Over the past few months, we’ve been pushing toward the frontier, step by step. And now, things are finally coming together. Welcome to StepFun. Come build better open models and more general agents with us!