StepFun AI released Step 5 Preview (open weight on October 15th.)
- 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
- 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
September 20, 2026 at 5:42 PM
StepFun AI released Step 5 Preview (open weight on October 15th.)
- 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
- 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
Step 3.5 Flash 🔥New MoE model from StepFun
huggingface.co/stepfun-ai/S...
huggingface.co/stepfun-ai/S...
✨ Base & Base-Midtrain
✨ 196B total/11B active - Apache 2.0
✨ 256K context
✨ High-speed reasoning & agentic tasks
huggingface.co/stepfun-ai/S...
huggingface.co/stepfun-ai/S...
✨ Base & Base-Midtrain
✨ 196B total/11B active - Apache 2.0
✨ 256K context
✨ High-speed reasoning & agentic tasks
stepfun-ai/Step-3.5-Flash-Base · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
March 3, 2026 at 12:54 PM
Step 3.5 Flash 🔥New MoE model from StepFun
huggingface.co/stepfun-ai/S...
huggingface.co/stepfun-ai/S...
✨ Base & Base-Midtrain
✨ 196B total/11B active - Apache 2.0
✨ 256K context
✨ High-speed reasoning & agentic tasks
huggingface.co/stepfun-ai/S...
huggingface.co/stepfun-ai/S...
✨ Base & Base-Midtrain
✨ 196B total/11B active - Apache 2.0
✨ 256K context
✨ High-speed reasoning & agentic tasks
Stepfun AI's Step-3.5.-Flash (open-weight and impressive benchmark)
A sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token.
Blog: static.stepfun.com/blog/step-3....
Model: huggingface.co/stepfun-ai/S...
A sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token.
Blog: static.stepfun.com/blog/step-3....
Model: huggingface.co/stepfun-ai/S...
February 2, 2026 at 5:21 AM
Stepfun AI's Step-3.5.-Flash (open-weight and impressive benchmark)
A sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token.
Blog: static.stepfun.com/blog/step-3....
Model: huggingface.co/stepfun-ai/S...
A sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token.
Blog: static.stepfun.com/blog/step-3....
Model: huggingface.co/stepfun-ai/S...
So 2025 wrapped.
*Trained SOTA small models.
*Published open datasets reused by Nvidia, Anthropic, Stepfun, Jina… and maybe more importantly, many independent researchers.
*Proved full synthetic training is viable in small size range (I have little doubts it will scale).
*Shipped a few blogposts
*Trained SOTA small models.
*Published open datasets reused by Nvidia, Anthropic, Stepfun, Jina… and maybe more importantly, many independent researchers.
*Proved full synthetic training is viable in small size range (I have little doubts it will scale).
*Shipped a few blogposts
December 31, 2025 at 5:32 PM
So 2025 wrapped.
*Trained SOTA small models.
*Published open datasets reused by Nvidia, Anthropic, Stepfun, Jina… and maybe more importantly, many independent researchers.
*Proved full synthetic training is viable in small size range (I have little doubts it will scale).
*Shipped a few blogposts
*Trained SOTA small models.
*Published open datasets reused by Nvidia, Anthropic, Stepfun, Jina… and maybe more importantly, many independent researchers.
*Proved full synthetic training is viable in small size range (I have little doubts it will scale).
*Shipped a few blogposts
Step Code
It runs in your terminal and handles the full task loop—reading code, making changes, and running tests. It works with the Step provider and discovers the available Step models after sign-in. MCP servers, Agent Skills, plugins, and multi-agent orchestration
github.com/stepfun-ai/S...
It runs in your terminal and handles the full task loop—reading code, making changes, and running tests. It works with the Step provider and discovers the available Step models after sign-in. MCP servers, Agent Skills, plugins, and multi-agent orchestration
github.com/stepfun-ai/S...
September 22, 2026 at 4:58 PM
Step Code
It runs in your terminal and handles the full task loop—reading code, making changes, and running tests. It works with the Step provider and discovers the available Step models after sign-in. MCP servers, Agent Skills, plugins, and multi-agent orchestration
github.com/stepfun-ai/S...
It runs in your terminal and handles the full task loop—reading code, making changes, and running tests. It works with the Step provider and discovers the available Step models after sign-in. MCP servers, Agent Skills, plugins, and multi-agent orchestration
github.com/stepfun-ai/S...
If you ever wondered if SYNTH could be usable in mid-training for larger models: Step-DeepResearch (from StepFun) is now out as private beta. stepfun.ai/deep-researc...
January 25, 2026 at 9:11 AM
If you ever wondered if SYNTH could be usable in mid-training for larger models: Step-DeepResearch (from StepFun) is now out as private beta. stepfun.ai/deep-researc...
StepFun AI released the SFT training dataset for Step-3.5-Flash-SFT.
Nice!
Dataset: huggingface.co/datasets/ste...
Project: static.stepfun.com/blog/step-3....
Nice!
Dataset: huggingface.co/datasets/ste...
Project: static.stepfun.com/blog/step-3....
stepfun-ai/Step-3.5-Flash-SFT · Datasets at Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
March 14, 2026 at 10:07 PM
StepFun AI released the SFT training dataset for Step-3.5-Flash-SFT.
Nice!
Dataset: huggingface.co/datasets/ste...
Project: static.stepfun.com/blog/step-3....
Nice!
Dataset: huggingface.co/datasets/ste...
Project: static.stepfun.com/blog/step-3....
Step-Video-T2V
A SOTA text-to-video pre-trained model with 30B parameters, capable of generating videos up to 204 frames long.
Models:
huggingface.co/stepfun-ai/s...
and
huggingface.co/stepfun-ai/s...
A SOTA text-to-video pre-trained model with 30B parameters, capable of generating videos up to 204 frames long.
Models:
huggingface.co/stepfun-ai/s...
and
huggingface.co/stepfun-ai/s...
February 17, 2025 at 6:03 PM
Step-Video-T2V
A SOTA text-to-video pre-trained model with 30B parameters, capable of generating videos up to 204 frames long.
Models:
huggingface.co/stepfun-ai/s...
and
huggingface.co/stepfun-ai/s...
A SOTA text-to-video pre-trained model with 30B parameters, capable of generating videos up to 204 frames long.
Models:
huggingface.co/stepfun-ai/s...
and
huggingface.co/stepfun-ai/s...
Step-Audio 2🔥 New end to end multimodal LLM for audio & speech, released by StepFun
huggingface.co/collections/...
huggingface.co/collections/...
Step-Audio 2 - a stepfun-ai Collection
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
August 29, 2025 at 8:45 AM
Step-Audio 2🔥 New end to end multimodal LLM for audio & speech, released by StepFun
huggingface.co/collections/...
huggingface.co/collections/...
StepFun's Step-Audio-R1 (open-weight)
- Test-Time Compute Scaling
- Deep Audio Comprehension
- Real-time responsiveness
- Scalable chain-of-thought reasoning for audio tasks
Comparable to Gemini 3 across major audio reasoning tasks.
huggingface.co/stepfun-ai/S...
- Test-Time Compute Scaling
- Deep Audio Comprehension
- Real-time responsiveness
- Scalable chain-of-thought reasoning for audio tasks
Comparable to Gemini 3 across major audio reasoning tasks.
huggingface.co/stepfun-ai/S...
stepfun-ai/Step-Audio-R1 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
November 28, 2025 at 11:09 PM
StepFun's Step-Audio-R1 (open-weight)
- Test-Time Compute Scaling
- Deep Audio Comprehension
- Real-time responsiveness
- Scalable chain-of-thought reasoning for audio tasks
Comparable to Gemini 3 across major audio reasoning tasks.
huggingface.co/stepfun-ai/S...
- Test-Time Compute Scaling
- Deep Audio Comprehension
- Real-time responsiveness
- Scalable chain-of-thought reasoning for audio tasks
Comparable to Gemini 3 across major audio reasoning tasks.
huggingface.co/stepfun-ai/S...
The DeepSeek competitors in China racing to make the next big breakthrough in AI, defying US export controls
Startups like Moonshot, Zhipu, MiniMax, StepFun, 01 ai and Baichuan are focusing on improving compute efficiency
tinyurl.com/5exjeb97
@bloomberg.com
#AI
Startups like Moonshot, Zhipu, MiniMax, StepFun, 01 ai and Baichuan are focusing on improving compute efficiency
tinyurl.com/5exjeb97
@bloomberg.com
#AI
February 6, 2025 at 3:41 AM
The DeepSeek competitors in China racing to make the next big breakthrough in AI, defying US export controls
Startups like Moonshot, Zhipu, MiniMax, StepFun, 01 ai and Baichuan are focusing on improving compute efficiency
tinyurl.com/5exjeb97
@bloomberg.com
#AI
Startups like Moonshot, Zhipu, MiniMax, StepFun, 01 ai and Baichuan are focusing on improving compute efficiency
tinyurl.com/5exjeb97
@bloomberg.com
#AI
StepFun released Step-Audio 2 mini - an open-weight 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
- 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
September 1, 2025 at 12:42 AM
StepFun released Step-Audio 2 mini - an open-weight 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
- 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
GitHub: github.com/stepfun-ai/S...
HuggingFace: huggingface.co/stepfun-ai/S...
GGUF: huggingface.co/stepfun-ai/S...
ModelScope: modelscope.cn/models/stepf...
API: platform.stepfun.ai
Blog: static.stepfun.com/blog/step-3....
HuggingFace: huggingface.co/stepfun-ai/S...
GGUF: huggingface.co/stepfun-ai/S...
ModelScope: modelscope.cn/models/stepf...
API: platform.stepfun.ai
Blog: static.stepfun.com/blog/step-3....
GitHub - stepfun-ai/Step-3.7-Flash
Contribute to stepfun-ai/Step-3.7-Flash development by creating an account on GitHub.
github.com
May 30, 2026 at 5:19 PM
GitHub: github.com/stepfun-ai/S...
HuggingFace: huggingface.co/stepfun-ai/S...
GGUF: huggingface.co/stepfun-ai/S...
ModelScope: modelscope.cn/models/stepf...
API: platform.stepfun.ai
Blog: static.stepfun.com/blog/step-3....
HuggingFace: huggingface.co/stepfun-ai/S...
GGUF: huggingface.co/stepfun-ai/S...
ModelScope: modelscope.cn/models/stepf...
API: platform.stepfun.ai
Blog: static.stepfun.com/blog/step-3....
More than a year after its initial release, Common Corpus has become an actual AI infrastructure:nearing a million downloads on HuggingFace, subsequent reuse by Nvidia, Anthropic, StepFun, ElasticSearch and most recently accepted as an ICRL oral.
February 19, 2026 at 12:48 PM
More than a year after its initial release, Common Corpus has become an actual AI infrastructure:nearing a million downloads on HuggingFace, subsequent reuse by Nvidia, Anthropic, StepFun, ElasticSearch and most recently accepted as an ICRL oral.
StepFun Says Step 3.7 Flash Matches 97% of Claude Opus 4.6’s Coding Performance at One-Ninth the Cost
firethering.com/stepfun-step...
#stepfun #claude #opus #coding #ai #technews #tech #llm #opensource
firethering.com/stepfun-step...
#stepfun #claude #opus #coding #ai #technews #tech #llm #opensource
StepFun Says Step 3.7 Flash Matches 97% of Claude Opus 4.6's Coding Performance at One-Ninth the Cost - Firethering
$0.19 vs $1.76. That's the per-task cost of running Step 3.7 Flash with Advisor Mode enabled versus Claude Opus 4.6 on SWE-Bench Verified. The Flash model scores 76.3% to Opus 4.6's 78.7%. Two percent...
firethering.com
May 30, 2026 at 5:03 PM
StepFun Says Step 3.7 Flash Matches 97% of Claude Opus 4.6’s Coding Performance at One-Ninth the Cost
firethering.com/stepfun-step...
#stepfun #claude #opus #coding #ai #technews #tech #llm #opensource
firethering.com/stepfun-step...
#stepfun #claude #opus #coding #ai #technews #tech #llm #opensource
StepFun AI's Step 3.7 Flash (open-weight)
400 TPS. 198B sparse MoE, ~11B active. 256K context, 3 reasoning levels, 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, built for agentic, coding, search, and multimodal workflows — balancing speed, cost, and reliable execution.
May 30, 2026 at 5:19 PM
StepFun AI's Step 3.7 Flash (open-weight)
400 TPS. 198B sparse MoE, ~11B active. 256K context, 3 reasoning levels, 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, built for agentic, coding, search, and multimodal workflows — balancing speed, cost, and reliable execution.
Chinese AI startup StepFun to unwind offshore structure to pave way for IPO, sources say reut.rs/3QyyrRc
Chinese AI startup StepFun to unwind offshore structure to pave way for IPO, sources say
Chinese AI agent StepFun is unwinding its offshore incorporation structure to pave the way for a planned Hong Kong initial public offering, three sources said, as Beijing tightens scrutiny of a system widely used for overseas fundraising.
reut.rs
April 13, 2026 at 7:00 AM
Chinese AI startup StepFun to unwind offshore structure to pave way for IPO, sources say reut.rs/3QyyrRc
Hear am sum reasoned pickshures of Shortie from mine Unkol Stepfun.
(Here are some recent pictures of Eddie, née ShortJorts, from my Uncle Stefan.)
(Here are some recent pictures of Eddie, née ShortJorts, from my Uncle Stefan.)
April 3, 2026 at 10:22 PM
Hear am sum reasoned pickshures of Shortie from mine Unkol Stepfun.
(Here are some recent pictures of Eddie, née ShortJorts, from my Uncle Stefan.)
(Here are some recent pictures of Eddie, née ShortJorts, from my Uncle Stefan.)
The NSA, FBI, and Cybersecurity and Infrastructure Security Agency allege that DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI have been attacking US models since at least late 2024.
Six Chinese AI firms accused of aggressively copying US frontier models
US urges AI firms to ID, then secretly switch, Chinese users to less-capable models.
arstechnica.com
September 10, 2026 at 5:23 PM
The NSA, FBI, and Cybersecurity and Infrastructure Security Agency allege that DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI have been attacking US models since at least late 2024.
After a VLM, StepFun dropped a new audio model: Step-Audio-R1.1, enabling thinking while speaking 🔥
huggingface.co/stepfun-ai/S...
✨ Apache 2.0
✨ Combines dual-brain architecture and acoustic-grounded reasoning to enable real-time dialogue with SOTA-level reasoning
huggingface.co/stepfun-ai/S...
✨ Apache 2.0
✨ Combines dual-brain architecture and acoustic-grounded reasoning to enable real-time dialogue with SOTA-level reasoning
stepfun-ai/Step-Audio-R1.1 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
January 16, 2026 at 4:16 PM
After a VLM, StepFun dropped a new audio model: Step-Audio-R1.1, enabling thinking while speaking 🔥
huggingface.co/stepfun-ai/S...
✨ Apache 2.0
✨ Combines dual-brain architecture and acoustic-grounded reasoning to enable real-time dialogue with SOTA-level reasoning
huggingface.co/stepfun-ai/S...
✨ Apache 2.0
✨ Combines dual-brain architecture and acoustic-grounded reasoning to enable real-time dialogue with SOTA-level reasoning