Written by Jakub Rusinowski · Last updated September 6, 2026
Model library → VibeThinker → VibeThinker 3B
A 3B dense reasoner built on Qwen2.5-Coder-3B. Reports 94.3 on AIME26, 89.3 on HMMT25 and 76.4 on IMO-AnswerBench — figures that put it in frontier-reasoner territory on competition maths while staying under 3 GB at Q4_K_M. Those numbers are the authors' own and remain disputed; the model is worth running precisely because you can check them yourself on a laptop.
VibeThinker 3B needs about 3 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
| Parameters | 3.09 Billion |
| Context window | 131,072 |
| Architecture | Dense Transformer (Qwen2.5-Coder-3B base) |
| Provider | WeiboAI |
| Licence | MIT |
| Specified at | Q4_K_M |
| System RAM | 8 GB |
| Record updated | 2026-09-06 |
MIT — commercial use permitted. Commercial use permitted. No usage restrictions beyond attribution.
Modelled on a reference NVIDIA RTX 4090 (24 GB), with no KV cache (this record has no published architecture). Speed figures are ESTIMATES from the memory-bandwidth roofline described on the methodology page, not benchmarks we ran — rows marked measured come from published or reader-submitted runs. VRAM here includes the KV cache, so it reads higher than the headline figure above, which does not.
| Quant | Weights | VRAM needed | Est. speed | Fit on 24 GB |
|---|---|---|---|---|
| Q2_K | 1.0 GB | 1.8 GB | ~241 tok/s (est.) | Fits comfortably |
| Q3_K_M | 1.3 GB | 2.1 GB | ~220 tok/s (est.) | Fits comfortably |
| Q4_K_M | 1.9 GB | 2.7 GB | ~190 tok/s (est.) | Fits comfortably |
| Q5_K_M | 2.2 GB | 3.0 GB | ~176 tok/s (est.) | Fits comfortably |
| Q6_K | 2.5 GB | 3.3 GB | ~163 tok/s (est.) | Fits comfortably |
| Q8_0 | 3.3 GB | 4.1 GB | ~140 tok/s (est.) | Fits comfortably |
| F16 | 6.2 GB | 7.0 GB | ~91 tok/s (est.) | Fits comfortably |
Want the memory numbers alone, at every quantization level and your own context length? Use the VibeThinker 3B VRAM calculator.
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
The cheapest catalogued GPU that runs VibeThinker 3B is the Intel Arc B570 (10 GB).
Install Ollama, then run:
ollama run vibethinkerWeights on Hugging Face: WeiboAI/VibeThinker-3B.
Quality scores as published by the model's authors or an independent evaluator — not throughput, and not measured by us.
| Benchmark | Score | Provenance |
|---|---|---|
| AIME 2026 | 94.3 / 100 % | vendor-claimed · https://huggingface.co/WeiboAI/VibeThinker-3B |
| HMMT 2025 | 89.3 / 100 % | vendor-claimed · https://huggingface.co/WeiboAI/VibeThinker-3B |
| IMO-AnswerBench | 76.4 / 100 % | vendor-claimed · https://huggingface.co/WeiboAI/VibeThinker-3B |
Best for: reasoning, math, research, edge devices.
← All VibeThinker models | VRAM calculator | Check your own hardware