GLM-4.5: Agentic, Reasoning, and Coding (Arc) Foundation Models [pdf]

50 SerCe 4 8/12/2025, 1:26:18 AM arxiv.org ↗

Comments (4)

darknoon · 8m ago
It's ok, somewhere between a qwen 2.5 VL and the frontier models (o3 / opus 4) on visual reasoning
ttul · 40m 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 · 37m ago
Why was qwen3 omitted from the coding benchmark but not other benchmarks?
Reubend · 15m ago
Fantastic release, and it's under the Apache license too. I'm so happy that we've got open source models pushing the envelope.