GLM-4.5: Agentic, Reasoning, and Coding (Arc) Foundation Models [pdf]
29 SerCe 2 8/12/2025, 1:26:18 AM arxiv.org ↗
Comments (2)
ttul · 13m ago
This feels like the first open model that doesn’t require significant caveats when comparing to frontier proprietary models. The parameter efficiency alone suggests some genuine innovations in training methodology. I am keen to see some independent verification of the results and to see how if does on Aider’s LLM Leaderboard.
lumost · 11m ago
Why was qwen3 omitted from the coding benchmark but not other benchmarks?