The story so far
Weights, every dated article, and repo commits on one rail.
12 repo commits since the weights landed
Teal is the weights (2025-05-29), grey ticks are repo commits.
Models › Qwen3-Reranker family
Qwen3-Reranker-0.6B
Qwen Apache 2.0 out 17 months ago
596M parameters, a query and passages in, relevance scores out, 40K tokens of context. 2 providers serve it from $0.010 per million input tokens.
Verified 2026-10-25 Rank #14 in Embeddings · score 102.2 method v1 Sources 3 Downloads 30d 1.6M · #78 of 619 tracked
Do people use it?1.6Mdownloads, 30 days#78 of 619 we track · top 13%+7 likes, downloads −3% this week424 likesHugging Face2026-10-24
Where to run it
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
Where it's used
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
Served by 2 API providers; heavily run locally (89 quantized builds).
- As a service2DeepInfra, HF Inference
- On own hardware89quantized builds
- Built on4733 fine-tunes · 14 adapters · 0 merges
- In apps33Hugging Face Spaces
Top Spaces running it: garage-lab/MCP_HTML2JSON ♥ 33 · kaengreg/rusBEIR ♥ 6 · sergeyzh/rusBEIR-lite ♥ 3 · ReallyFloppyPenguin/Qwen3-Reranker-0.6B ♥ 2
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 0.4 GB at INT4, 0.7 GB at FP8, 1.4 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
What people built on it
136 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
136 derivatives on the Hub, most of them quantized
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
What the card says
1 cited sentence
Use-case evidence
8 use cases
What it is good at 8 use cases
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
- Rerank search resultsstructuraldecision-model estimate p=1.00name: Reranktag: text-ranking+1 more
- Run on a laptop or phonestructuraldecision-model estimate p=0.99name: 0.6Bfits: edge
- Call tools and APIsstructuralchat template: <tool_call>
- Reason through hard problemsstructuralchat template: <think>
- Chat and follow instructionsstructuralchat template present
- Embed text for searchstructuraltag: sentence-transformers
- Classify and make typed decisionsstructuraldecision-model estimate p=0.80
- Work across languagesstructuraldecision-model estimate p=1.00
Facts
21 fields
Listed in Embeddings
Facts
- Qwen3-Reranker-0.6B has 596M parameters. (safetensors metadata, observed 2026-10-24)
- Context length is 40,960 tokens (40K). (config.json max_position_embeddings, observed 2026-10-24)
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
- Created 2025-05-29, last modified 2026-04-16. (HF API, observed 2026-10-24)
| Pipeline tag | text-ranking |
|---|---|
| Library | transformers |
| Architectures | Qwen3ForCausalLM |
| Model type | qwen3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 595,776,512 |
| Size band | S |
| MoE | no |
| Context length | 40,960 (config.json max_position_embeddings) |
| Papers | arXiv:2506.05176 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · Qwen3ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3ForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Hardware fit
3 estimates
Hardware fit ESTIMATE
Weights-only memory for Qwen3-Reranker-0.6B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.4 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.7 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.4 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
40 variants
Lineage
Base model: Qwen/Qwen3-0.6B-Base
quantized: 89finetune: 33adapter: 14
Top variants (40 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| ggml-org/Qwen3-Reranker-0.6B-Q8_0-GGUF | quantized | Q8_0 | — | — | apache-2.0 | ggml-org | 29 | 458k |
| dolfsai/Qwen3-Reranker-0.6B-seq-cls-vllm-W8A8 | quantized | -W8A8 | 596M | S | apache-2.0 | dolfsai | 0 | 142.3k |
| mradermacher/Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 9 | 86.3k |
| Voodisss/Qwen3-Reranker-0.6B-GGUF-llama_cpp | quantized | GGUF | — | — | apache-2.0 | Voodisss | 12 | 35.7k |
| pipizhao/SkillRouter-Reranker-0.6B | finetune | — | 596M | S | apache-2.0 | pipizhao | 9 | 32.2k |
| DevQuasar/Qwen.Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | — | DevQuasar | 9 | 14.5k |
| furiosa-ai/Qwen3-Reranker-0.6B | finetune | — | 596M | S | apache-2.0 | furiosa-ai | 0 | 8.7k |
| mku64/Qwen3-Reranker-0.6B-mlx-8Bit | quantized | mlx | 596M | S | apache-2.0 | mku64 | 1 | 8.5k |
| drawais/Qwen3-Reranker-0.6B-AWQ-INT4 | quantized | AWQ | 599M | S | apache-2.0 | drawais | 2 | 8.1k |
| n24q02m/Qwen3-Reranker-0.6B-ONNX | quantized | — | — | — | apache-2.0 | n24q02m | 3 | 7.3k |
| thebajajra/RexReranker-0.6B | finetune | — | 596M | S | apache-2.0 | thebajajra | 10 | 6.1k |
| mlx-community/Qwen3-Reranker-0.6B-4bit | quantized | -4bit | 596M | S | apache-2.0 | mlx-community | 4 | 6.4k |
| tencent/R3-rerank-0.6b | finetune | — | 596M | S | apache-2.0 | tencent | 32 | 4.5k |
| boboliu/Qwen3-Reranker-0.6B-W4A16-G128 | quantized | -W4A16 | 599M | S | apache-2.0 | boboliu | 2 | 4.6k |
| onnx-community/Qwen3-Reranker-0.6B-ONNX | quantized | — | — | — | apache-2.0 | onnx-community | 3 | 4.5k |
| ayanami-kitasan/code-pruner | finetune | — | 634M | S | mit | ayanami-kitasan | 11 | 3.9k |
| giladgd/Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | apache-2.0 | giladgd | 0 | 4.4k |
| shawnw3i/Qwen3-Reranker-0.6B-ONNX | quantized | — | — | — | apache-2.0 | shawnw3i | 10 | 3k |
| OpenVINO/Qwen3-Reranker-0.6B-fp16-ov | finetune | -fp16 | — | — | apache-2.0 | OpenVINO | 0 | 2.6k |
| paulalesius/Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | apache-2.0 | paulalesius | 1 | 2.3k |
| mradermacher/Qwen3-Reranker-0.6B-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 1 | 2.2k |
| Rnfudge/snapd-reranker-v1 | finetune | — | 596M | S | gpl-3.0 | Rnfudge | 0 | 2.3k |
| dean2155/Qwen3-Reranker-0.6B-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | dean2155 | 0 | 2k |
| tensorblock/Qwen_Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | apache-2.0 | tensorblock | 1 | 1.8k |
| JamePeng2023/Qwen3-Reranker-GGUF | quantized | GGUF | — | — | mit | JamePeng2023 | 0 | 1.9k |
| n24q02m/Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | apache-2.0 | n24q02m | 2 | 1.7k |
| Nicolassuez/Qwen3-Reranker-0.6B-W8A16-g128-gptq-resident | finetune | gptq | 599M | S | apache-2.0 | Nicolassuez | 0 | 1.8k |
| dean2155/Qwen3-Reranker-0.6B-Q8_0-GGUF | quantized | Q8_0 | — | — | apache-2.0 | dean2155 | 0 | 1.5k |
| Hanno-Labs/bosun-xs | adapter | — | — | — | apache-2.0 | Hanno-Labs | 3 | 1.3k |
| arthurcollet/Qwen3-Reranker-0.6B-mlx-6bit | quantized | mlx | 596M | S | apache-2.0 | arthurcollet | 0 | 1.4k |
| AIPlans/Qwen3-HHH-Cipher-Eng | finetune | — | 596M | S | — | AIPlans | 0 | 1.4k |
| seroe/Qwen3-Reranker-0.6B-turkish-triplet | finetune | — | 596M | S | apache-2.0 | seroe | 0 | 1.3k |
| DuyTa/sec-rerank | finetune | — | 596M | S | apache-2.0 | DuyTa | 0 | 1.2k |
| Mithilss/Qwen3-Reranker-0.6B-finetune-lower-lr | finetune | — | 596M | S | apache-2.0 | Mithilss | 0 | 1.2k |
| Mithilss/Qwen3-Reranker-0.6B-finetune | finetune | — | 596M | S | apache-2.0 | Mithilss | 0 | 1.2k |
| batiai/Qwen3-Reranker-0.6B-GGUF | quantized | GGUF | — | — | apache-2.0 | batiai | 1 | 1k |
| futurist-ai/Qwen3-Reranker-0.6B-MLX-4bit | quantized | MLX | 596M | S | apache-2.0 | futurist-ai | 0 | 999 |
| tatsuyaaaaaaa/Qwen3-Reranker-0.6B-gguf | quantized | gguf | — | — | apache-2.0 | tatsuyaaaaaaa | 0 | 994 |
| samdotci/Qwen3-Reranker-0.6B-mlx-4Bit | quantized | mlx | 596M | S | apache-2.0 | samdotci | 1 | 814 |
| Hanno-Labs/bosun-xs-GGUF | quantized | GGUF | — | — | apache-2.0 | Hanno-Labs | 0 | 769 |
Repository changelog
12 commits
Repository changelog 12 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-04-16 | Integrate with Sentence Transformers v5.4 (#24) | e61197e |
| 2025-06-09 | update model type and add paper link (#5) | 6e9e698 |
| 2025-06-05 | Update README.md | ad4c588 |
| 2025-06-05 | Update README.md | ea4c160 |
| 2025-06-05 | Update README.md | 5340c02 |
| 2025-06-05 | Delete special_tokens_map.json | adcdb2a |
| 2025-06-05 | Delete added_tokens.json | 3861be7 |
| 2025-06-05 | Update README.md | 9748b7a |
| 2025-06-05 | Update README.md | fd9fb1d |
| 2025-05-29 | Update README.md | 204631f |
| 2025-05-29 | Upload folder using huggingface_hub | c97ad57 |
| 2025-05-29 | initial commit | 0b048c5 |
All providers
2 providers · 2 endpoints
Where to run 2 providers · 2 endpoints
Hosted endpoints observed for Qwen3-Reranker-0.6B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.01/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepInfra deepinfra | Qwen/Qwen3-Reranker-0.6B | 32K | — | $0.01 | — | — | — | — | live | 2026-10-24 |
| HF Inference hf | Qwen/Qwen3-Reranker-0.6B | — | — | — | — | — | — | — | live | 2026-10-25 |
Articles and coverage
4 pieces
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Analysis4
dev.co887 words
Qwen3-Reranker-0.6B is a 595M parameter lightweight text reranking model from Alibaba's Qwen team designed to score and reorder candidate documents for relevance to a query. It handles 32k-token contexts, supports 100+ languages, and is Apache 2.0 licensed for production deployment. Flash Attention 2 compatibility reduces memory footprint and improves throughput during batch inference.written, reviewed, cited
- The model card does not explicitly document fine-tuning, LoRA, or QLoRA support.
- Running the model requires transformers>=4.51.0 for the qwen3 model class.
- Some organizations may have vendor or geopolitical constraints regarding the Chinese-origin model.
reranker.ukabout the Qwen3 Reranker family762 words
Qwen3-Reranker offers Apache 2.0 rerankers in 0.6B, 4B, and 8B sizes with a 32K context and 100+ languages. The 4B and 8B variants lead on MTEB-R and FollowIR suites across English, Chinese, multilingual, long-document, and code domains. Engineers should validate on their own labels before replacing bge or hosted APIs, noting that GPU resources are required for comfortable operation.written, reviewed, cited
- The 8B variant leads on Chinese, multilingual, long-document, and code tasks by at most 1.5 points.
- The raw transformers path requires Qwen's exact prompt template, otherwise quality drops without an error.
- The 4B model achieves the best FollowIR score in Qwen's table among the available sizes.
savrn.com1,687 words
Qwen3-Reranker-0.6B is a 596M-parameter cross-encoder released under Apache License 2.0 for AI agent skill retrieval. The model supports a 40,960-token context and fits in a single 1.2 GB safetensors file, requiring about 1.4 GB of GPU memory at 16-bit precision. It is fine-tuned from Qwen/Qwen3-0.6B-Base and described in arXiv:2506.05176.written, reviewed, cited
deepinfra.com1,681 words
DeepInfra offers the Qwen3-Reranker-0.6B model for text reranking tasks. It supports over 100 languages and has a context length of 32k. The model requires transformers>=4.51.0 and vllm>=0.8.5 for inference. Using instructions typically yields an improvement of 1% to 5% compared to not using them.written, reviewed, cited
- The 'flex' service tier processes requests at lower priority for a discount but may be retried or timed out under load.
- In multilingual contexts, users are advised to write instructions in English because most training instructions were originally written in English.
- Using Transformers versions earlier than 4.51.0 results in a KeyError: 'qwen3' error during inference.
Papers
1 paper
Papers 1
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Similar models
3 models
Similar models
- jina-reranker-v2-base-multilingual · 278M same size band (S), same task (text-ranking)
- Qwen3-Embedding-0.6B · 596M same size band (S), fewer providers, permissive license
- multilingual-e5-large · 560M same size band (S), permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25
Follow Qwen3-Reranker-0.6B
Watch this model
- /feeds/models/Qwen--Qwen3-Reranker-0.6B.xml(RSS 2.0) · every change to Qwen3-Reranker-0.6B
- /feeds/families/Qwen--Qwen3-Reranker.xml(RSS 2.0) · every change across the Qwen3-Reranker family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.