| Phi 4 | Kimi K2.6 | |
|---|---|---|
| Input, per 1M tokens | $0.07 | $0.65 |
| Output, per 1M tokens | $0.14 | $3.41 |
| Context window | 16K | 262K |
| Classification or routing (600 tokens in, 10 out), per 1,000 | $0.0434 | $0.424 |
| Chat reply (2,000 in, 400 out), per 1,000 | $0.196 | $2.66 |
| Retrieval-augmented answer (8,000 in, 500 out), per 1,000 | $0.63 | $6.91 |
| Confidence signal | no | yes |
| Tool calling | no | yes |
| BANKING77, correct of 400 | not measured | 323 |
Which one should carry your traffic?
If Phi 4 answers your requests as well as Kimi K2.6, moving that traffic saves 93% on chat-shaped requests. Averages cannot tell you whether it does for your requests. Phi 4 does not expose a confidence signal through OpenRouter, so Gimbix would test it as a full replacement rather than a first answer.
Phi 4 profile · Kimi K2.6 profile · Measure it on your traffic