Qwen3-Reranker 0.6B
Alibaba · Reranker · 0.6B · 32k context · Released 5 June 2025
A reranker does not retrieve; it refines. After an embedding search returns a rough shortlist, Qwen3-Reranker reads each candidate passage together with the query and scores true relevance, so the best few can be kept and the rest dropped. Adding this step is one of the highest-value improvements to a local RAG system, because it fixes the common case where the right passage was retrieved but ranked too low to be used. At 0.6B parameters it adds little latency and runs comfortably on modest hardware, or on the CPU. Larger 4B and 8B siblings exist for higher quality at more cost.
Strengths
- Sharpens RAG results, often the single biggest retrieval-quality win
- Multilingual, covering many languages
- Tiny and Apache 2.0, so it runs anywhere with no commercial conditions
Weaknesses
- Adds a little latency per query, since every candidate is scored
- Needs a first-stage retriever, so it is an addition to a pipeline, not a replacement
- Benchmark figures below are the developer's own, on reranking-specific tests
Hardware requirements
| Quantisation | Approx. VRAM | Notes |
|---|---|---|
| FP16 | ~2GB | A 0.6B model is tiny, so full precision is the norm and fits anything |
Also runs on CPU (slower). Optimised builds available for Apple Silicon.
What you'd need to run this
Roughly what a machine to run this would need, at up to three levels of quality. Memory is the deciding factor.
Minimum to run it
FP16 · ~2GB needed
One 12GB GPU
NVIDIA GeForce RTX 3060 12GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 12GB runs →Licence
Apache 2.0 — read the licence
Benchmarks
| Benchmark | Score | Source | As of |
|---|---|---|---|
| MTEB-R | 65.80 | Qwen3-Reranker model card | June 2025 |
| MMTEB-R | 66.36 | Qwen3-Reranker model card | June 2025 |
| MLDR | 67.28 | Qwen3-Reranker model card | June 2025 |
How it compares
How this model’s reported scores sit against other models we cover, on the same benchmarks. This model is highlighted.
MTEB-R
higher is better- Qwen3-Reranker 0.6B 65.80
Qwen3-Reranker model card · June 2025
- BGE-reranker-v2-m3 57.03
Qwen3-Reranker card comparison table (competitor-measured) · June 2025
Qwen3-Reranker 0.6B: common questions
- What hardware do I need to run Qwen3-Reranker 0.6B?
- At its most compressed (FP16) it needs roughly 2GB of VRAM, and about 2GB for good quality. VRAM figures are approximate and depend on context length and settings.
- Is Qwen3-Reranker 0.6B free for commercial use?
- Yes. Qwen3-Reranker 0.6B is licensed under Apache 2.0, which permits commercial use with no meaningful conditions.
- Can I run Qwen3-Reranker 0.6B on Apple Silicon?
- Yes. Qwen3-Reranker 0.6B has builds optimised for Apple Silicon, through MLX or GGUF on a Mac.
- Does Qwen3-Reranker 0.6B run on CPU?
- Yes, Qwen3-Reranker 0.6B can run on the CPU, though generation is slower than on a GPU.
- What is Qwen3-Reranker 0.6B's context window?
- Qwen3-Reranker 0.6B has a context window of 32,000 tokens, about 32k.
Availability
Recommended for
- Improving retrieval quality in a local RAG system
- Multilingual reranking on modest hardware
- Pairing with a local embedding model like BGE-M3 or Qwen3-Embedding
Related models
Run it with
Related guides
Catalogue entry last verified 18 August 2026. Specifications change; verify anything you are about to spend money on.