Qwen3-Max-Thinking Outperforms Claude Opus 4.5 and Gpt-5.2: The Panic Is Real

Alibaba’s Qwen3 family of models has become a major alternative in the LLM ecosystem.

In April 2025, the Qwen team released Qwen3, an open‑weight suite containing dense and mixture‑of‑experts (MoE) models ranging from 0.6 billion to 235 billion parameters.

These models introduced a hybrid thinking design, enabling step‑by‑step reasoning when needed and fast responses otherwise.

Qwen3 family supports 119 languages and dialects, and up to 128 000 token context lengths, making it attractive for multilingual and long‑context applications.

Qwen3-Max-Thinking that is introduced this week is the family’s latest flagship model.

It’s trained with massive scale and advanced RL, it delivers strong performance across reasoning, knowledge, tool use, and agent capabilities:

• Adaptive tool-use which leverages Search, Memory & Code Interpreter without manual selection
• Test-time scaling for multi-round self-reflection, which beats Gemini 3 Pro on reasoning
• From complex math (98.0 on HMMT Feb) to agentic search (49.8 on HLE)

On 19 established benchmarks, it demonstrates performance comparable to leading models such as GPT-5.2-Thinking, Claude-Opus-4.5, and Gemini 3 Pro.