GLM 5.5 in post-training? Lumina says the three sizes are not three releases
A new claim points to internal 32B, 122B and 256B models. The author says those numbers describe training work, not a downloadable lineup.
GLM 5.5 is already in post-training, according to Lumina (@LuminaBench). In an October 31 post, the account describes internal 32B, 122B and 256B models that it believes will contribute to the next base model. That is a claim about development, not an announcement that new weights are available.
The original GLM 5.5 claim
Loading post from X…
The author’s post, not an official model announcement. Its attached artifact has not been independently authenticated.
The distinction matters if you are waiting for a smaller model to run locally. In a follow-up, Lumina explicitly says the three sizes are for post-training, not three separate models users should expect to receive. The author refers to a commit, but we have not independently located and authenticated that artifact. The proposed merging process remains the author’s interpretation.
There is no launch date in the inspected posts. In another reply, Lumina says it does not know how far away a release is. Neither the training claim nor the parameter counts establish coding performance, hardware requirements or public availability. For now, the lead is a possible next GLM model—not a confirmed release schedule.
Sources
- GLM 5.5 post-training claimLumina / X
- Clarification: the three sizes are for post-trainingLumina / X
- The author says release timing is unknownLumina / X
Earliest source found: . Updated: .
