| Model | Score | Status | Released | Price in/out | Context | |
|---|---|---|---|---|---|---|
| Qwen2.5-72B-Instruct mature multilingual open-weight workhorse | 7.5 | GA | 2024-09-18 | $0.12 / $0.30 | 131K | Review → |
| Qwen2.5-32B-Instruct mature Apache-2.0 single-GPU workhorse | 7.2 | GA | 2024-09-18 | $0.10 / $0.25 | 131K | Review → |
Qwen2.5-72B-Instruct was Alibaba's open-weight flagship from late 2024 until the Qwen3 release in April 2025, and remains in heavy production use. It is a 72.7B-parameter dense decoder that competes with Llama 3.1 70B and, on several benchmarks, with Llama 3.1 405B. The buyer's sentence: a mature, dependable, broadly multilingual open weight with the largest community fine-tune ecosystem after Llama — keep it if you have it, but start new builds on Qwen3-32B. - Provider: Alibaba Cloud (Qwen Team) - Released: 2024-09-19 (GA) - Tier: Large dense - Context: 131,072 tokens (YaRN; 32K native) - Max output: 8,192 tokens - Modalities: text in, text out - Knowledge cutoff: approx. 2024-06 - Headline price: $0.12 in / $0.30 out per 1M tokens (typical blended)
Full review →