GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
qiusizhan
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
qiusizhan
Model Tree
Qwen/Qwen3.5-9B
Input Modalities
Output Modalities
Supported Functionality
The exported arm completed 2 fresh-base optimizer updates. A separate capacity arm completed four updates padded to 5,120 positions before being stopped early; its weights are not this checkpoint. That arm peaked at 145.89 GiB allocated and 168.22 GiB reserved. Both arms stayed within one original 45-minute / $6.80 estimate; actual billing was not measured. Pod stop was independently confirmed.
All eight held-out inputs produced diagnostic token sequences. Only three
reached native EOS; five reached the 128-token limit. Incomplete generations
are not negative ASR results. No final ASR/FPR, trigger selectivity or benchmark
readiness is claimed. Training throughput was slow and its cause remains
unresolved. Exact software/source/model hashes and measured steps are in
smoke-record.json; no private held-out texts or action targets are published.