Yuchen Jin stepped in to answer a direct question from another researcher about how recent side-by-side impressions of GLM-5.2 were actually run, naming the OpenCode harness paired with a custom Databricks endpoint for the model. The exchange followed an initial post noting that the open-source model sometimes matched or edged out GPT-5.5 and Opus 4.8 in personal use, prompting the call for provider and tooling specifics.
@Yuchenj_UW In which provider+harness did you try it?
After using GLM-5.2 for a day, I’m surprised by how often it feels close to Opus 4.8/GPT-5.5 level.
I compared it side by side with Opus 4.8, and sometimes I even preferred GLM-5.2’s results.
OSS LLMs are impressive, especially given how many fewer GPUs they were trained on.
OpenCode is one of several open coding-agent evaluation tools whose exact implementation can swing results, so spelling out both the harness and the non-public Databricks endpoint removes one layer of guesswork for anyone trying to read the informal comparison.
The thread supplies the tooling details but stops short of quantitative pass rates, deltas, or full methodology tables, leaving the strength of the GLM-5.2 claims dependent on the original anecdotal observations rather than reproducible figures.
No Digg Deeper questions have been answered for this story yet.
@giffmana I used OpenCode + our own Databricks GLM-5.2 endpoint.
@Yuchenj_UW In which provider+harness did you try it?
@giffmana @Yuchenj_UW http://SidekickPro.com