Inf-Retriever-v1
7B-class bilingual embedding that topped the AIR-Bench leaderboard in early 2025, fine-tuned from gte-Qwen2-7B-instruct.
About
Inf-Retriever-v1 is the flagship embedding model from Infiniflow (publishing as infly on Hugging Face), released in late 2024 as a fine-tune of Alibaba's gte-Qwen2-7B-instruct. It targets Chinese and English retrieval with a 32K-token input window, making it one of the few long-context bilingual embedding models with fully open Apache 2.0 weights.
The model card reports an average score of 52.56 on the AIR-Bench heterogeneous retrieval leaderboard, which it topped as of January 2025, beating its own base model and BGE-Multilingual-Gemma2 across wiki, web, healthcare, law, and news splits. A smaller 1.5B variant (inf-retriever-v1-1.5b) has seen far wider adoption than the 8B flagship, which suggests the headline model is best for those who can afford the compute.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- AIR-Bench average: 52.56, leaderboard-top as of Jan 23, 2025 (author-reported, model card)
- Beats gte-Qwen2-7B-instruct (48.38) and BGE-Multilingual-Gemma2 (46.83) on the same board (author-reported)
- Full per-task model-index scores published in the model card
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (verified tier, metrics-v1). Evidence: HF API confirms Apache 2.0 license and Dec 2024 release; AIR-Bench 52.56 figure taken from the authors' own model card table. Strengths: genuinely open license, strong author-reported multilingual scores, long context. Docked for: (1) only ~750 HF downloads on the 8B flagship, so real-world validation is thin; (2) no peer-reviewed paper, all figures author-reported; (3) the 1.5B variant is the one the community actually uses.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 4.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · HF API confirms Apache 2.0 and Dec 2024 release; 7B-class fine-tune of gte-Qwen2-7B-instruct, 32K context
- · AIR-Bench 52.56 leaderboard-top (Jan 2025) is author-reported in the model card
- · Only ~750 HF downloads on the 8B flagship; the 1.5B variant has ~24K downloads
- · No peer-reviewed paper; all figures author-reported