DeepSeek: DeepSeek V4 Flash vs. StepFun: Step 3.7 Flash
Direkter Head-to-Head-Vergleich zweier Frontier-Modelle. DeepSeek: DeepSeek V4 Flash gewinnt 4 von 8 Disziplinen.
Letzte Synchronisation:
|
DeepSeek: DeepSeek V4 Flash
DeepSeek
Mehr von DeepSeek →
|
StepFun: Step 3.7 Flash
Stepfun
Mehr von Stepfun →
|
|
|---|---|---|
| Quality Index | 46 ★ | 42,6 |
| Speed (Tokens/s) | 0 | 403,1 ★ |
| Latency (TTFT) | 0 ms ★ | 786 ms |
| Preis Input (USD/1M) | $0.14 ★ | $0.20 |
| Preis Output (USD/1M) | $0.28 ★ | $1.15 |
| Context Window | — | — |
| Modalitäten | text | text |
| Release | 04/2026 | 05/2026 |
DeepSeek: DeepSeek V4 Flash
DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window. It is designed for fast inference and...
StepFun: Step 3.7 Flash
Step 3.7 Flash is StepFun's latest high-efficiency multimodal Mixture-of-Experts model. It pairs a 196B-parameter language backbone with a vision encoder for native image and video understanding, activating roughly 11B parameters...