StepFun releases Step 3.7 Flash with vision and auto-escalation

StepFun has released Step 3.7 Flash, an upgraded version of its 3.5 Flash model, featuring a new vision encoder and an automatic "Advisor Mode" that escalates complex tasks to larger models. This update aims to improve performance consistency, narrowing the variance in output quality across different tasks. The model is compatible with OpenAI's SDK, with a key change being the strict requirement for an exact regional base URL to avoid silent errors. AI

IMPACT This release offers improved performance consistency and new features like native vision processing, potentially impacting the development of coding agents and other AI applications.

RANK_REASON This is a new model release from a frontier lab (StepFun) with a specific name and version. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• 3.5 flash
• Advisor Mode
• Claude Opus-4.6
• NVIDIA NIM
• OpenAI
• OpenRouter
• Step-3.7-Flash
• StepFun
• SWE Bench Pro
• Terminal-Bench 2.1

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →