Contextual AI Reranker v2 2B
Release in the Contextual AI Reranker family · version ctxl-rerank-v2-instruct-multilingual-2b
Maintained by Contextual AI1
The 2B-parameter size of Contextual AI Reranker v2, an instruction-following multilingual reranker released in August 2025. Given a query, an optional natural-language instruction, and a document, it outputs a relevance score; the card lists support for more than 100 languages and inputs of up to 32K tokens.14
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Release notes: Announcement (external site: contextual.ai)
- License: CC BY-NC-SA 4.0 license (external site: creativecommons.org)
Availability and license
Overall availability
Downloadable from Hugging Face without gating, for noncommercial use under CC BY-NC-SA 4.0. Contextual AI also serves managed versions through its hosted Rerank API.134
Availability is separate from permission: read the license before using or redistributing.
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (external site: creativecommons.org)1
CC BY-NC-SA 4.0 prohibits commercial use and requires derivatives to be shared under the same license, so the weights are open for research and other noncommercial use only.1
Component reuse rights
- weights
- Terms or license combinations require review
- code
- Unknown — no complete fact-level rights review
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | BF16 weights are on Hugging Face; the announcement also links NVFP4 quantized versions published as separate repositories.14 |
| Inference codeIs code for running the model published? | Public | The card gives inference code for Sentence Transformers (CrossEncoder), vLLM, and Hugging Face Transformers, including the prompt template and how scores are read from the logits.1 |
| Training codeIs the code used to train the model published? | Unknown | No training code is linked from the card or announcement. |
| Training-data informationDoes the information cover provenance, scope, acquisition, selection, labeling, processing, and where data or alternatives can be obtained? Access alone does not establish completeness. | Unknown | The card and announcement reviewed do not describe the training data. |
| Training-data accessCan the training data be obtained? This is independent of information completeness and reuse rights; original unshareable data need not be downloadable. | Unknown | Not assessed. |
| Complete training pipelineIs the complete base-training and preprocessing pipeline published, including configuration? Fine-tuning code or an inference SDK alone is insufficient. | Unknown | Not assessed. |
| Legacy data assessment (v0.1)Historical assessment combining download access and disclosure. Preserved for traceability; excluded from the v0.2 tier calculation. See the new separate assessments above. | Unknown | Not assessed. |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Unknown | The card and announcement reviewed do not describe the training method. |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The card and announcement report benchmark results (including on BEIR) as charts; no evaluation code is linked.14 |
What it is useful for
Re-ranking retrieved documents with custom instructions, such as prioritizing recent information, handling conflicting retrieval results, and multilingual or long-context search.1
Run and use notes
- The card recommends vLLM for production (vLLM 0.8.5 or later for BF16, vLLM 0.10.0 for NVFP4) and documents a simpler Hugging Face Transformers path (Transformers 4.51.0 or later, BF16 only; NVFP4 is not supported there).1
Organization context
Provenance and derivatives
The card and announcement do not name a base checkpoint. The published configuration declares the Qwen3ForCausalLM architecture (model type qwen3), which is the architecture of Alibaba Cloud's Qwen3 models; whether the weights were initialized from a Qwen3 checkpoint is not documented in the sources reviewed.216
Other releases in the Contextual AI Reranker family
No other releases in this family have been assessed.
What this catalog does not know
- Training code: unknown.
- Training-data information: unknown.
- Training-data access: unknown.
- Complete training pipeline: unknown.
- Legacy data assessment (v0.1): unknown.
- Training recipe: unknown.
Have a primary source? How to report a correction.
U.S. eligibility
Eligible · basis: U.S. headquarters
The model is published in Contextual AI's Hugging Face organization and the card names Contextual AI as its developer. Contextual AI, Inc. gives a Mountain View, California postal address in its privacy notice (see the Contextual AI record). The checkpoint the model was trained from is not disclosed; its configuration declares the Qwen3 architecture (see provenance), so this record does not treat any underlying base weights as U.S.-developed.152
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.