DEV Community

WDSEGA
WDSEGA

Posted on • Originally published at wdsega.github.io

Step 3.7 Flash: 416 tokens/s, 1/9 the Cost of Claude, 97% of Its Coding Ability

Chinese AI startup Stepfun just released Step 3.7 Flash, and the benchmarks are turning heads globally.

Key Numbers

Metric Value
Max output speed 416 tokens/s
Cost vs Claude Opus 1/9
Coding ability vs Claude ~97%
Artificial Analysis speed rank #1
Artificial Analysis value rank #1

Why This Matters

For most real-world tasks - code review, documentation, Q&A, content generation - you do not need the absolute best model. You need one that is fast, accurate enough, and cheap.

Step 3.7 Flash hits 97% of Claude's coding ability at 1/9 the price. For high-volume API use cases, this translates directly to your monthly bill.

Who Should Try It

  • High-frequency API callers
  • Real-time applications needing low latency
  • Cost-sensitive production deployments
  • Chinese language tasks

Full analysis: wdsega.github.io

Top comments (0)