| Model | Score | Status | Released | Price in/out | Context | |
|---|---|---|---|---|---|---|
| QwQ-32B open-weight always-on reasoning at 32B | 6.8 | GA | 2025-03-04 | $0.12 / $0.18 | 131K | Review → |
QwQ-32B is Alibaba's open-weight reasoning model — the direct response to DeepSeek-R1 and OpenAI o1/o3-mini — shipped to full GA 2025-03-05 under Apache 2.0 (a preview shipped November 2024). It is a 32.5B dense decoder trained with reinforcement learning to produce long chain-of-thought by default; unlike Qwen3's optional thinking toggle, QwQ-32B is always reasoning. The buyer's sentence: DeepSeek-R1-class reasoning at 32B parameters, single-GPU and Apache-licensed, but always-on CoT makes it a routed sub-tier, not a general default. - Provider: Alibaba Cloud (Qwen Team) - Released: 2025-03-05 (GA); QwQ-32B-Preview shipped 2024-11-28 - Tier: Reasoning specialist - Context: 131,072 tokens (32K native + YaRN) - Max output: 32,768 tokens (reasoning chains are long) - Modalities: text in, text out - Knowledge cutoff: approx. 2024-09 - Headline price: approx. $0.12 in / $0.18 out per 1M tokens (DeepInfra)
Full review →