Models

We host these ourselves. Speeds are measured on our own hardware at a single stream, not quoted from a datasheet.

Model	tokens/s	Context	Input / 1M	Output / 1M
`glm-4.7-flash` Fast, tool calls, logprobs, 202k context. The default: the lower-priced of the two and the one most workloads should use.	62.8	202,752	0.25	1.50
`qwen3.8-27b` 27.3B dense at F16, 262k context, tool calls, logprobs. Slower per token, priced 3x the flash model because it holds the GPUs 3.6x longer.	17.5	262,144	0.75	4.50

Prices exclude 5% UAE VAT, which applies to UAE customers.