StepFun · MoE Architecture
Step 3.7 Flash
198B-parameter MoE vision-language model pairing a 196B language backbone with a 1.8B vision encoder — activates ~11B params/token, 256k context, three reasoning levels, ~400 tok/s throughput. With Advisor Mode hits 97% of Claude Opus 4.6's SWE-bench Verified score at ~1/9 the per-task cost ($0.19 vs $1.76).
Open the source ↗Similar Language Models releases
- Groot-280B-A16B-261023
GrootTeam · Open reasoning MoE · Week 45, 2026
- Apex-35B-V1.0
Apex Intelligence · Research automation LLM · Week 45, 2026
- Ling-3.0-flash-Sante
inclusionAI (Ant Group) · Medical LLM · Week 44, 2026
- Kimi K3.1 Pro
Moonshot AI · Open-Weight Frontier · Week 44, 2026
- DeepSeek-V4.1-Flash-1017
DeepSeek · Open-Weight LLM · Week 44, 2026
- LFM2.5-24B-A2B
Liquid AI · Efficient Agentic MoE · Week 43, 2026
More from StepFun
- Step 5 Preview
StepFun · Open Weight Release · Week 43, 2026
- ACE-Step 1.5
StepFun · Music Generation · Week 15, 2026
- Step-3.5-Flash
StepFun · Reasoning · Week 6, 2026
- Step-Audio-R1.1
StepFun · Voice Agents · Week 4, 2026
- Step3-VL-10B
StepFun · Visual Understanding · Week 4, 2026
- Step-Audio-EditX
StepFun · Text-to-Speech · Week 46, 2025
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.