| Claude Sonnet 5 | GLM 5.3 | |
|---|---|---|
| Input, per 1M tokens | $2.00 | $0.325 |
| Output, per 1M tokens | $10.00 | $4.40 |
| Context window | 1M | 1.31M |
| Classification or routing (600 tokens in, 10 out), per 1,000 | $1.30 | $0.239 |
| Chat reply (2,000 in, 400 out), per 1,000 | $8.00 | $2.41 |
| Retrieval-augmented answer (8,000 in, 500 out), per 1,000 | $21.00 | $4.80 |
| Confidence signal | no | yes |
| Tool calling | yes | yes |
| BANKING77, correct of 400 | 308 | not measured |
Which one should carry your traffic?
If GLM 5.3 answers your requests as well as Claude Sonnet 5, moving that traffic saves 70% on chat-shaped requests. Averages cannot tell you whether it does for your requests. GLM 5.3 exposes a confidence signal, so Gimbix can let it answer first and hand unsure cases to Claude Sonnet 5.
Claude Sonnet 5 profile · GLM 5.3 profile · Measure it on your traffic