Step 5 Preview (High): benchmark results

Provider: StepFun. Released 2026-09-20. Access: API.

Aggregate Rating 1778 ± 23, rank #195 of 2245 rated models, from 72 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
FrontierCS 2 - OCR Distillation Trajectory Sampling85.45Task score (0–100)94.7
SEAL - SWE Atlas - Codebase QnA63.6Score90.6
FrontierCS 2 - EVM State Checkpoints66.92Task score (0–100)89.5
SEAL - MCP Atlas85.6Score88.7
AA CritPt20.9Accuracy (%)88.6
FrontierCS 2 - In-place Integer Redistribution51.07Task score (0–100)78.9
WebDev Arena (Brand & Marketing)1590Arena Score78.4
WebDev Arena1570.15Arena Score78
WebDev Arena (Gaming)1608Arena Score77
WebDev Arena (Frontend)1576Arena Score75.7
Chatbot Arena (Code)1564Elo75.2
WebDev Arena (Simulations)1571Arena Score74.1

Interactive version: theaggregate.ai/model?slug=step-5-preview-high · How It Works · Data refreshed daily, snapshot 2026-11-16.