| MiniMax M3 | Qwen3.8 Flash | |
|---|---|---|
| Input, per 1M tokens | $0.30 | $0.15 |
| Output, per 1M tokens | $1.20 | $0.47 |
| Context window | 1.05M | 1M |
| Classification or routing (600 tokens in, 10 out), per 1,000 | $0.192 | $0.0947 |
| Chat reply (2,000 in, 400 out), per 1,000 | $1.08 | $0.488 |
| Retrieval-augmented answer (8,000 in, 500 out), per 1,000 | $3.00 | $1.44 |
| Confidence signal | yes | yes |
| Tool calling | yes | yes |
| BANKING77, correct of 400 | 316 | 316 |
Which one should carry your traffic?
If Qwen3.8 Flash answers your requests as well as MiniMax M3, moving that traffic saves 55% on chat-shaped requests. Averages cannot tell you whether it does for your requests. Qwen3.8 Flash exposes a confidence signal, so Gimbix can let it answer first and hand unsure cases to MiniMax M3.
MiniMax M3 profile · Qwen3.8 Flash profile · Measure it on your traffic