Daxin Jiang

[StepFun @StepFun_ai]

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: https://t.co/fC7HHlWKFn
Model page: https://t.co/4caJR2YGD3

Open weights on Oct 15.

[StepFun @StepFun_ai]

⚡️ Step 3.7 Flash is here: The new frontier is agent efficiency.

#1 ClawEval-1.1 (67.1), #1 SimpleVQA Search (79.2), #2 SWE-PRO (56.3), 95.3 on V* Python. Open weights under Apache 2.0.

Built for agentic, coding, search, and multimodal workflows — balancing speed, cost, and reliable execution.

- 400 TPS. 198B sparse MoE, ~11B active. 256K context, 3 reasoning levels.
- Understands UIs, charts, docs, images — then writes code or calls tools to act on what it sees.
- Web + visual search reaches further: more sources, deeper follow-up.
- Reliable tool use — less drift, fewer broken toolcalls. 98%+ on τ²-bench across all difficulty levels.
- Works with Claude Code, KiloCode, Hermes Agent, OpenClaw, and protocols like MCP.
- Runs locally on Mac Studio M4 Max, DGX Spark, AMD AI Max+ 395.

GitHub: https://t.co/kqlZkVIRHv
HuggingFace: https://t.co/qqceCrgPiw
GGUF: https://t.co/rR6XrnymWG
ModelScope: https://t.co/wney6Tzvqy
API: https://t.co/RvHWzRG7Fu
Blog: https://t.co/BxDiajiQ5G

[StepFun @StepFun_ai]

💥BIGGER IS BETTER!

🚀 Introducing Step-GUI：A larger, more robust GUI model.

Our trusty on-device model, GELab-Zero-4B, just got a new name — Step-GUI Edge. But that's just the start. Step-GUI takes things to the next level with:

🧠 Mobile to desktop, Continuous intelligence
🔍 Stronger semantic understanding
🌍 Better generalization across GUI tasks

And here's the kicker: with GUI-MCP, you can deploy in just 10 minutes! ⏱️

GitHub：https://t.co/lWlaUCgYH2

[StepFun @StepFun_ai]

🫣Is GPT-realtime too pricey for you? Here is a free alternative!!! 🤩Releasing Step-Audio 2 mini (7B): new SOTA open source LALM!🔥🔥🔥
✨ Free alternative to expensive GPT Realtime API
✨ End-to-end speech I/O
✨ Advanced speech & audio understanding
✨ Expressive prosody control
✨ Intelligent speech conversation
✨ Web search support
🏆 SOTA on LibriSpeech, MMAU, URO-Bench & more!
👇Check out and try:
🔗 GitHub: https://t.co/1yYbbj1zNs
🔗 Huggingface：https://t.co/oWjQDIskYg

[vLLM @vllm_project]

This amazing Attention-FFN disaggregation implementation from @StepFun_ai , achieves decoding throughput of up to 4,039 tokens per second per
GPU under 50ms TPOT SLA, for their 321B-A38B MoE model Step3 served with H800! The implementation is based on vLLM, and we are working together to bring it to the public! Kudos to @StepFun_ai 🚀 Check out their tech report at https://t.co/3RExMgM2zO .

[StepFun @StepFun_ai]

🚀 Announcing Step 3: Our latest open-source multimodal reasoning model is here! Get ready for a stronger, faster, & more cost-effective VLM！
🔵 321B parameters (38B active), optimized for top-tier performance & cost-effective decoding.
🔵 Revolutionary Multi-Matrix Factorization Attention (MFA) and Attention-FFN Disaggregation (AFD) enable efficient inference—even on modest GPUs.
🔵 Trained on 20T+ tokens (incl. 4T multimodal), with meticulous data curation ensuring reduced hallucinations & robust reasoning across vision and language.
🚄 Unmatched speed: Up to 4,039 tokens/sec/GPU—70% faster than DeepSeek-V3 under similar conditions.
💎 Step 3 sets a new Pareto frontier—bridging power, efficiency, and practicality.
👉 Start building with Step 3 today: https://t.co/S0EKQcmgh3
👉More details on our research blog：
https://t.co/kX1ls7O5SZ

[StepFun @StepFun_ai]

🚀 StepFun releases StepMesh, the powerful communication library tailored for Attention-FFN disaggregation serving systems! It is a key enabler driving Step-3's exceptional performance.🔥
🚄 High Performance: Ultra-low latency & high bandwidth without the hassle of barrier synchronization.
⚡ Asynchronous & SM-free: Seamlessly overlaps with GPU computation for peak efficiency.
🛠️ Tensor-Native APIs: Developer-friendly and built for ease of use.
👉 Dive into the details:
Code: https://t.co/r5U5Ptk63Z
Report: https://t.co/B55ZPXNHEr