GLM-5.1 and Qwen3 looked cheaper and faster in one benchmark
The benchmark sent the same inputs to several models and recorded cost, delay, and s from each vendor's live API usage response. It ran on ScitiX's , so the numbers should be treated as results from one specific setup. For text generation, GLM-5.1 cost $0.0007 per call and produced its first token after 706 milliseconds.
cost $0.0067 for the same task and produced its first token after 1051 milliseconds. For embeddings, Qwen3-Embedding-8B cost $0.04 per million tokens and took 311 milliseconds, while OpenAI text-embedding-3-large cost $0.13 per million tokens and took 1685 milliseconds. On the same English input, GLM-5.1 used 838 tokens while Sonnet 4.5 used 947 tokens, so GLM-5.1 used about 11.5% fewer tokens before pricing was even applied.
Turning on for GLM-5.1 raised token cost by about 9.3 times compared with leaving it off.
Key points
- GLM-5.1 was about 9 times cheaper than on the same generation task.
- Qwen3-Embedding-8B was cheaper per million tokens and faster than OpenAI text-embedding-3-large in this embedding test.
- Tokenizer density mattered: GLM-5.1 used fewer tokens for the same English input.
- can be expensive; for GLM-5.1 it raised token cost by about 9.3 times.
- Agent cost comparisons should include price, , delay, and mode-specific costs.