| DeepSeek V4 Flash 0731 | Phi 4 | |
|---|---|---|
| Input, per 1M tokens | $0.021 | $0.07 |
| Output, per 1M tokens | $0.32 | $0.14 |
| Context window | 1.31M | 16K |
| Classification or routing (600 tokens in, 10 out), per 1,000 | $0.0158 | $0.0434 |
| Chat reply (2,000 in, 400 out), per 1,000 | $0.17 | $0.196 |
| Retrieval-augmented answer (8,000 in, 500 out), per 1,000 | $0.328 | $0.63 |
| Confidence signal | yes | no |
| Tool calling | yes | no |
| BANKING77, correct of 400 | 298 | not measured |
Which one should carry your traffic?
If DeepSeek V4 Flash 0731 answers your requests as well as Phi 4, moving that traffic saves 13% on chat-shaped requests. Averages cannot tell you whether it does for your requests. DeepSeek V4 Flash 0731 exposes a confidence signal, so Gimbix can let it answer first and hand unsure cases to Phi 4.
DeepSeek V4 Flash 0731 profile · Phi 4 profile · Measure it on your traffic