VibeThinker-3B
3B parameter open-weight model. Built on Qwen2.5-3B with a 64K long-context training window focused on complete reasoning trajectories. No official Ollama library tag (registry checked 2026-07-06) — runs on consumer GPUs with 4–6 GB VRAM via llama.cpp or LM Studio using community GGUF quantizations. Not suited for tool-calling or agentic workflows.
WeiboAI · VibeThinker
Model overview
VibeThinker-3B is WeiboAI's compact reasoning-focused model, built on top of Qwen2.5-3B and trained with an emphasis on complete step-by-step reasoning trajectories rather than short, clipped answers. Its small size makes it one of the lightest reasoning-capable options in the catalog, running on 4-6 GB of VRAM at Q4 or Q8 through community GGUF builds in llama.cpp or LM Studio. It's a good fit if you want to experiment with math or logic-style reasoning on modest hardware, but it's explicitly not tuned for tool-calling or agentic workflows, so pick a different model if that's your primary use case.
Editorial review
VRAM figures are empirical estimates. Actual usage varies by runtime, context length, and system configuration. Verify on your specific hardware before production use.
Will VibeThinker-3B run on your machine?
VibeThinker-3B is 3B parameters and needs 3.9 GB of VRAM at Q4_K_M — 2.4 GB of weights plus 1.5 GB of runtime overhead for the inference server itself.
VRAM by quantization
| Quantization | Weights | Needs (with overhead) | Quality |
|---|---|---|---|
| Q4_K_M | 2.4 GB | 3.9 GB | good |
| Q8_0 | 3.8 GB | 5.3 GB | high |
Fit on common hardware at Q4_K_M
| Hardware | Memory the model can use | System RAM | Verdict |
|---|---|---|---|
| CPU Only | None (CPU only) | 16 GB | CPU offload |
| RTX 4060 Laptop | 8 GB | 16 GB | Comfortable |
| RTX 3060 (12GB) | 12 GB | 32 GB | Comfortable |
| RTX 4060 Ti (16GB) | 16 GB | 32 GB | Comfortable |
Comfortable means VRAM clears the requirement by 2 GB or more. Tight means it covers the requirement with no margin. CPU offload means the model does not fit in VRAM but system RAM is at least 1.6× the weights, so it will run at reduced speed — expect roughly 1–5 tokens per second. Figures are weights plus a fixed runtime overhead and exclude KV-cache growth, which scales with context length.
Need more hardware for VibeThinker-3B? Open the PC Builder for the 1B–3B lightweight tier →
This catalog has no verified local Ollama tag for this checkpoint, so fit grades do not include a local run command.
Ready to run this model locally?
Find a compatible interface in our Local AI Tools directory →