GLM-4.5: Agentic, Reasoning, and Coding (Arc) Foundation Models [pdf]
27 SerCe 2 8/12/2025, 1:26:18 AM arxiv.org ↗
Comments (2)
ttul · 5m 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 · 3m ago
Why was qwen3 omitted from the coding benchmark but not other benchmarks?