Alibaba

Qwen2.5

Alibaba's open-weight model family (0.5B–72B), with the flagship Qwen2.5-72B-Instruct competitive with much larger open models. Trained on ~18T tokens with strong multilingual and long-context support. Linked findings reflect documented failure-mode classes; per-version attribution is illustrative.

Strengths

Open weights across many sizes; strong multilingual coverage and long context; competitive reasoning and coding for its size class.

Known weaknesses

Shares the frontier reasoning, arithmetic and hallucination classes; base (non-instruct) variants are not tuned for conversation.

Findings (0)

No findings linked to this model's versions yet.

Versions tracked

qwen2.5-72b-instruct

Cite this

Qlarify Labs. (2026). Alibaba Qwen2.5 — known weaknesses. Retrieved from https://labs.qlarify.fi/models/qwen2.5