Zhipu AI / Z.ai GLM Watch posted an update
MiaAIlab says its GLM 5.3 Flash setup for four NVIDIA DGX Spark systems will offer two weight options: EXL3 by default, or the original FP8 weights. The EXL3 version is 164GB, compared with 306GB for FP8.
Why it mattersThe account expects similar prefill speeds but noticeably faster decoding with EXL3. It says quality evaluations of both options are underway, so that speed expectation is not yet a published test result. For builders planning local inference, the practical trade-off is already clear: EXL3 takes substantially less space, while the quality comparison is still to come.
Discuss: Would you choose the smaller, faster-expected option now, or wait for the quality evaluations before committing?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.