StepFun reports a substantial preview-model release with multimodal input and long-context support, but claimed capabilities remain unverified and promised open weights are not yet available.
Limits of the evidence: Single independent report; no independent evaluation is provided, and the source says open weights are scheduled for a future date.
What to watch
Assessment revised Oct 4, 2026
StepFun Launches Step 5 Preview: A 600B-Total, 27B-Active MoE Model With 1M Context for Long-Horizon Agentic Work
MarkTechPostSep 21, 2026First report
First assessment, from MarkTechPost
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.