GLM 5.3 FlashX API: the fast GLM 5.3 Flash, about 3x the speed
GLM 5.3 FlashX is live on FastMetal as glm-5.3-flashx. It is the high-speed version of GLM 5.3 Flash, served by Z.ai itself. The answers are much the same as Flash; what you pay for is output speed.
How fast it is
We sent the same prompt (a roughly 600-character explanation, written in Japanese) three times to each model on launch day:
| Model | Run 1 | Run 2 | Run 3 |
|---|---|---|---|
glm-5.3-flashx | 141 tok/s | 160 tok/s | 176 tok/s |
glm-5.3-flash | 42 tok/s | 113 tok/s | 91 tok/s |
That is about three times the throughput, and FlashX was also steadier from run to run. GLM 5.3 Flash has moved to different hosts since these runs, so its speed today may not match the second row.
What else it does
- A context window of about 1 million tokens, enough to pass a long document whole.
- Image input. Given a photo of a Japanese store receipt, it read the store name, the total and the items taxed at the reduced rate correctly, in about four seconds.
- Tool calling.
Answer quality matched Flash in our checks. On a task to fix mistakes in Japanese honorific speech, it found both errors and explained each one correctly.
Flash or FlashX
FlashX costs more per token than Flash. Both bill per million input tokens and per million output tokens; the current rates are on the GLM 5.3 FlashX and GLM 5.3 Flash model pages. Since the answers are about the same, the choice comes down to whether waiting costs you anything:
- A user is watching the answer arrive (a chat UI, a voice assistant): FlashX.
- Nobody is waiting (batch jobs, overnight summarization): Flash.
FlashX reasons before it answers. On longer answers, about 70% of the output tokens were reasoning tokens, and those bill as output. When a short answer is enough, lower the effort in the reasoning object to cut both latency and cost. The reasoning docs show how.
Calling it
curl https://api.fastmetal.ai/v1/chat/completions \
-H "Authorization: Bearer $FASTMETAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3-flashx",
"messages": [{"role": "user", "content": "Explain the difference between cloud and on-premises hosting in about 150 words."}]
}'
Moving from Flash is a one-line change: set model to glm-5.3-flashx. Your API key and balance stay the same.
FAQ
What is the difference between GLM 5.3 FlashX and GLM 5.3 Flash?
Speed and price. Answer quality is about the same; in our launch-day runs FlashX was about three times faster and it costs more per token.
Does GLM 5.3 FlashX accept images?
Yes. It read a Japanese receipt photo correctly in our test.
How is GLM 5.3 FlashX billed?
Per million input tokens and per million output tokens, from a prepaid balance. Rates are on the model page and in pricing.
To read the two models' answers side by side, open the GLM 5.3 Flash vs GLM 5.3 FlashX comparison.