Qwen’s Former Lead on What Hybrid Thinking Got Wrong — and Why He Now Backs Agents Junyang Lin, former technical lead of Alibaba's Qwen project, criticized hybrid thinking models in AI and advocated for a shift toward agent-based systems. In a talk and subsequent post, he detailed Qwen3's architecture and explained why merging thinking and non-thinking modes degrades performance, arguing that reasoning should be shaped by workload rather than benchmarks. Junyang Lin https://www.linkedin.com/in/junyang-lin-0b2b38151/ was the technical lead of Alibaba’s Qwen project. He announced he was stepping down on March 3, 2026 https://techcrunch.com/2026/03/03/alibabas-qwen-tech-lead-steps-down-after-major-ai-push/ . He now lists himself as an independent researcher on his personal site. In a talk titled ‘ Qwen: Towards a Generalist Model / Agent, https://www.youtube.com/watch?v=b0xlsQ 6wUQ ‘ he walks through the Qwen family. It ends on a single line: “Training models - training agents.” He later expanded that line into an detailed post as an independent researcher. This article reads the talk and the detailed post together. What Lin’s Talk Actually Covers The talk is a tour of the Qwen model family, not a single release. It moves through QwQ-32B, Qwen2.5-Max, Qwen3, Qwen2.5-VL, and Qwen2.5-Omni. Each stop shows benchmark charts against contemporaries. The named baselines include DeepSeek-R1, Grok 3 Beta, Gemini 2.5 Pro, and OpenAI’s o-series. The Qwen3 stop carries the most detail. Lin highlights hybrid thinking modes: a thinking mode for step-by-step reasoning, and a non-thinking mode for near-instant responses. He adds dynamic thinking budgets, so callers can cap how much the model reasons. Qwen3 expanded multilingual support from 29 to 119 languages and dialects. The presentation lists many model types and sizes from 0.6B to 235B parameters. It also lists quantized formats including GGUF, GPTQ, AWQ, and MLX, all under Apache 2.0. Two demos follow: a Web Dev demo and a Deep Research demo. The closing “Future work” slide points at agents. It lists more pretraining, RL with environment feedback, longer context, and more modalities. The last key mention is the “training models - training agents.” Qwen3 Architecture, As Shown in the Talk The talk includes the Qwen3 architecture tables, reproduced below. | Model | Layers | Heads Q/KV | Tie Embedding / Experts Total/Act. | Context | |---|---|---|---|---| | Qwen3-0.6B | 28 | 16 / 8 | Tie: Yes | 32K | | Qwen3-1.7B | 28 | 16 / 8 | Tie: Yes | 32K | | Qwen3-4B | 36 | 32 / 8 | Tie: Yes | 32K | | Qwen3-8B | 36 | 32 / 8 | Tie: No | 128K | | Qwen3-14B | 40 | 40 / 8 | Tie: No | 128K | | Qwen3-32B | 64 | 64 / 8 | Tie: No | 128K | | Qwen3-30B-A3B | 48 | 32 / 4 | Experts: 128 / 8 | 128K | | Qwen3-235B-A22B | 94 | 64 / 4 | Experts: 128 / 8 | 128K | The small dense models tie input and output embeddings and use a 32K context. The larger dense and MoE models drop tying and extend context to 128K. The two MoE models activate 8 of 128 experts per token. Hybrid Thinking, and Why Merging is Hard Lin presents hybrid thinking as a clean feature. The post explains why it was hard to build. Lin writes that thinking mode and instruct mode pull in opposite directions. A strong instruct model is rewarded for directness, brevity, and low latency. A strong thinking model is rewarded for spending more tokens on hard problems. Merge the two carelessly, and both degrade. The thinking behavior gets bloated, and the instruct behavior gets less crisp. Qwen3 tried the merge with a four-stage post-training pipeline. That pipeline included a long-CoT cold start, reasoning RL, and a “thinking mode fusion” step. Later in 2025, the 2507 line shipped separate Instruct and Thinking variants instead. Lin frames this as a data problem more than a model problem. Anthropic took the opposite route, and Lin calls it a useful corrective. Claude 3.7 Sonnet shipped as a hybrid model with a user-set thinking budget. Claude 4 let reasoning interleave with tool use, aimed at coding and long-running tasks. His point: a longer reasoning trace does not make a model smarter. Thinking should be shaped by the target workload, not by the benchmark. Interactive Explainer From ‘Reasoning’ Thinking to ‘Agentic’ Thinking Lin draws a line between two eras. The first was reasoning thinking, defined by o1 and DeepSeek-R1. It taught the field that RL needs deterministic, verifiable rewards, so math, code, and logic became central. It also turned RL into a systems problem of large-scale rollouts and verification. The next era, in his framing, is agentic thinking: thinking in order to act. An agent formulates plans, decides when to act, uses tools, reads environment feedback, and revises. It is defined by closed-loop interaction with the world, not by a long internal monologue. Lin lists what agentic thinking must handle that pure reasoning can avoid: - Deciding when to stop thinking and take an action - Choosing which tool to invoke, and in what order - Incorporating noisy or partial observations from the environment - Revising plans after failures - Maintaining coherence across many turns and many tool calls The optimization target changes with the era. The table below summarizes the contrast Lin draws. | Dimension | Reasoning thinking | Agentic thinking | |---|---|---| | Judged by | Quality of internal deliberation before an answer | Whether progress is sustained while acting | | Reward signal | Verifiable answers math, code, logic | Task success in an interactive environment | | Core object of training | The model | The model plus its environment the harness | | Infra bottleneck | Rollouts, verification, stable policy updates | Tool servers, sandboxes, train-serve decoupling | | Main failure mode | Verbose, low-value reasoning traces | Reward hacking through tool access and env leaks | Use Cases, With Examples The distinction changes how you build: Coding agents : A reasoning model emits one patch from a stack trace. An agentic system runs the test harness, reads the real error, revises, and re-runs until the suite passes. Thinking here should help with codebase navigation, error recovery, and tool orchestration. Deep research : A reasoning model writes a long answer from memory. An agentic system breaks the question into sub-queries, calls search, drops weak sources, and returns grounded citations. Qwen’s own Deep Research demo sits in this category. Multi-agent orchestration : Lin expects ‘harness engineering’ to matter more. An orchestrator plans and routes work. Specialized sub-agents execute narrower tasks and help control context pollution. A Concrete Hook: Qwen3 Thinking Toggle Hybrid thinking is exposed directly in code. The enable thinking flag switches modes in the chat template. python from transformers import AutoModelForCausalLM, AutoTokenizer name = "Qwen/Qwen3-8B" tok = AutoTokenizer.from pretrained name model = AutoModelForCausalLM.from pretrained name, torch dtype="auto", device map="auto" messages = {"role": "user", "content": "Refactor this function and explain the change."} enable thinking=True - step-by-step thinking mode enable thinking=False - near-instant, non-thinking mode text = tok.apply chat template messages, tokenize=False, add generation prompt=True, enable thinking=True, inputs = tok text, return tensors="pt" .to model.device Qwen's recommended sampling for thinking mode out = model.generate inputs, max new tokens=2048, temperature=0.6, top p=0.95, top k=20, enable thinking=True is the default, and the output wraps reasoning in a