{"ok":true,"source":"tensorfeed.ai","lastUpdated":"2026-10-19","count":37,"models":[{"id":"text-embedding-3-large","name":"text-embedding-3-large","provider":"OpenAI","type":"embedding","dimensions":3072,"maxInputTokens":8192,"pricePer1MTokens":0.13,"pricingNote":"$0.13 per 1M input tokens. Reducible-dimensions: pass `dimensions` to truncate to 256/1024/etc with minor quality loss.","openSource":false,"license":"Proprietary","released":"2024-01-25","notes":"OpenAI flagship embedding and still its newest; no successor has shipped. Strong on English retrieval; supports Matryoshka truncation to lower dimensions for cheaper storage. Azure lists a 2028-02-09 retirement for its hosted copy.","multilingual":true,"url":"https://developers.openai.com/api/docs/guides/embeddings","mtebAvg":64.6},{"id":"text-embedding-3-small","name":"text-embedding-3-small","provider":"OpenAI","type":"embedding","dimensions":1536,"maxInputTokens":8192,"pricePer1MTokens":0.02,"pricingNote":"$0.02 per 1M input tokens. The default budget choice for English RAG.","openSource":false,"license":"Proprietary","released":"2024-01-25","notes":"OpenAI budget embedding. 5x cheaper than ada-002 with better quality. Good default for most RAG agents.","multilingual":true,"url":"https://developers.openai.com/api/docs/guides/embeddings","mtebAvg":62.3},{"id":"voyage-4-large","name":"voyage-4-large","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.12,"pricingNote":"$0.12 per 1M tokens after 200M free tokens per account. Batch API is 33% cheaper. Output dimensions 256, 512, 1024 (default), or 2048.","openSource":false,"license":"Proprietary","released":"2026-01-15","notes":"Voyage flagship on a mixture-of-experts architecture. Every voyage-4 model writes into the same embedding space, so documents indexed with voyage-4-large can be queried with the cheaper voyage-4-lite. A third cheaper than voyage-3-large.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":null},{"id":"voyage-4","name":"voyage-4","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.06,"pricingNote":"$0.06 per 1M tokens after 200M free tokens per account.","openSource":false,"license":"Proprietary","released":"2026-01-15","notes":"Mid-tier voyage-4 model and the current Voyage workhorse. Shares an embedding space with the rest of the voyage-4 series.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":null},{"id":"voyage-4-lite","name":"voyage-4-lite","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.02,"pricingNote":"$0.02 per 1M tokens after 200M free tokens per account. Up to 1M tokens per batch request.","openSource":false,"license":"Proprietary","released":"2026-01-15","notes":"Latency and cost tier of the voyage-4 series. The shared embedding space makes it a natural query-side model against a corpus indexed with voyage-4-large.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":null},{"id":"voyage-code-4","name":"voyage-code-4","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.12,"pricingNote":"$0.12 per 1M tokens, a third below voyage-code-3. Output dimensions 256, 512, 1024 (default), or 2048.","openSource":false,"license":"Proprietary","released":"2026-08-13","notes":"Code retrieval model tuned for coding-agent workloads. Voyage reports a 27.54% gain over voyage-code-3 on agentic retrieval and double-digit leads over Cohere Embed 4 and Gemini Embedding 2 on code search.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":null},{"id":"voyage-4-nano","name":"voyage-4-nano","provider":"Voyage AI","type":"embedding","dimensions":2048,"maxInputTokens":32000,"pricePer1MTokens":null,"pricingNote":"Open weights on Hugging Face. Free to self-host.","openSource":true,"license":"Apache-2.0","released":"2026-01-15","notes":"The open-weight member of the voyage-4 series: 180M non-embedding plus 160M embedding parameters. Voyage says all voyage-4 embeddings are compatible with each other, so a self-hosted nano index can later be queried with the hosted voyage-4 models.","multilingual":true,"url":"https://huggingface.co/voyageai/voyage-4-nano","mtebAvg":null},{"id":"voyage-3-large","name":"voyage-3-large","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.18,"pricingNote":"$0.18 per 1M input tokens. Listed under older models, with no free-token allowance.","openSource":false,"license":"Proprietary","released":"2025-01-07","notes":"Previous-generation Voyage flagship. voyage-4-large replaces it at $0.12, so new projects have no price reason to start here. Supports Matryoshka truncation.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":67},{"id":"voyage-3","name":"voyage-3","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.06,"pricingNote":"$0.06 per 1M input tokens. Listed under older models.","openSource":false,"license":"Proprietary","released":"2024-09-18","notes":"Older Voyage workhorse, superseded by voyage-3.5 and then voyage-4 at the same price. Keep only for existing indexes.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":63.5},{"id":"voyage-3-lite","name":"voyage-3-lite","provider":"Voyage AI","type":"embedding","dimensions":512,"maxInputTokens":32000,"pricePer1MTokens":0.02,"pricingNote":"$0.02 per 1M input tokens. Listed under older models.","openSource":false,"license":"Proprietary","released":"2024-09-18","notes":"Older budget tier. 512-dim vectors keep storage costs low; voyage-4-lite is the current option at the same price.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":60},{"id":"voyage-code-3","name":"voyage-code-3","provider":"Voyage AI","type":"embedding","dimensions":1024,"maxInputTokens":32000,"pricePer1MTokens":0.18,"pricingNote":"$0.18 per 1M input tokens. Listed under older models.","openSource":false,"license":"Proprietary","released":"2024-12-05","notes":"Previous code-search model, replaced by voyage-code-4 at a lower price.","multilingual":true,"url":"https://docs.voyageai.com/docs/embeddings","mtebAvg":null},{"id":"embed-v5.0-pro","name":"embed-v5.0-pro","provider":"Cohere","type":"embedding","dimensions":2048,"maxInputTokens":128000,"pricePer1MTokens":0.12,"pricingNote":"$0.12 per 1M text tokens, $0.40 per 1M image tokens. Output dimensions 256, 512, 768, 1024, 1536, or 2048 (default), in float32, int8, or binary.","openSource":false,"license":"Proprietary","released":"2026-09-30","notes":"Cohere's new flagship embedding, replacing embed-v4.0 as the recommended model. Pro and Fast share one vector space, so Cohere's own advice is to index with Pro and query with Fast. Text, images, and mixed documents such as PDFs in one space, 100+ languages, 128K context. Vendor figures: ViDoRe V3 average 85.8 against 83.7 for voyage-4-large and 83.2 for gemini-embedding-2, and 82.3 on a five-dataset fused text-image average against 61.3 for gemini-embedding-2.","multilingual":true,"url":"https://docs.cohere.com/docs/cohere-embed","mtebAvg":null},{"id":"embed-v5.0-fast","name":"embed-v5.0-fast","provider":"Cohere","type":"embedding","dimensions":2048,"maxInputTokens":128000,"pricePer1MTokens":0.08,"pricingNote":"$0.08 per 1M text tokens, $0.40 per 1M image tokens (the same image rate as Pro). Same six output dimensions as Pro.","openSource":false,"license":"Proprietary","released":"2026-09-30","notes":"Latency and cost tier of Embed 5, a third cheaper than Pro on text. The shared vector space is the point: a corpus indexed with Pro can be queried with Fast without re-embedding. Vendor figures put it 1.3 points behind Pro on the ViDoRe V3 average (84.5 to 85.8) and 1.1 behind on fused text-image.","multilingual":true,"url":"https://docs.cohere.com/docs/cohere-embed","mtebAvg":null},{"id":"embed-v4.0","name":"embed-v4.0","provider":"Cohere","type":"embedding","dimensions":1536,"maxInputTokens":128000,"pricePer1MTokens":0.12,"pricingNote":"$0.12 per 1M text tokens, $0.47 per 1M image tokens. Output dimensions 256, 512, 1024, or 1536 (default).","openSource":false,"license":"Proprietary","released":"2025-04-15","notes":"Previous Cohere flagship, superseded by Embed 5 Pro and Fast on 2026-09-30 at the same and a lower price. Still served, and still the only v4 model Cohere documents with an embedding price. Embeds text, images, and mixed documents such as PDFs into one space, 100+ languages, and a 128K context that removes most chunking pressure. Embed 5 adds a 2048-dimension option and int8 and binary output formats, which v4 does not list.","multilingual":true,"url":"https://docs.cohere.com/docs/cohere-embed","mtebAvg":null},{"id":"embed-multilingual-v3.0","name":"embed-multilingual-v3.0","provider":"Cohere","type":"embedding","dimensions":1024,"maxInputTokens":512,"pricePer1MTokens":0.1,"pricingNote":"$0.10 per 1M input tokens at last listing. Cohere's public pricing page now shows only Embed 4, so confirm the v3 rate in the dashboard.","openSource":false,"license":"Proprietary","released":"2023-11-02","notes":"Previous Cohere multilingual model, still served. 100+ languages with strong cross-lingual retrieval, but the 512-token input limit is the main constraint next to embed-v4.0.","multilingual":true,"url":"https://docs.cohere.com/docs/cohere-embed","mtebAvg":64},{"id":"embed-english-v3.0","name":"embed-english-v3.0","provider":"Cohere","type":"embedding","dimensions":1024,"maxInputTokens":512,"pricePer1MTokens":0.1,"pricingNote":"$0.10 per 1M input tokens at last listing. Not shown on the current public pricing page.","openSource":false,"license":"Proprietary","released":"2023-11-02","notes":"English-only sibling of embed-multilingual-v3. Slightly stronger on English-only corpora.","multilingual":false,"url":"https://docs.cohere.com/docs/cohere-embed","mtebAvg":64.5},{"id":"gemini-embedding-2","name":"gemini-embedding-2","provider":"Google","type":"embedding","dimensions":3072,"maxInputTokens":8192,"pricePer1MTokens":0.2,"pricingNote":"$0.20 per 1M text tokens on the Gemini API and Vertex AI; images $0.45, audio $6.50, video $12.00 per 1M tokens. Batch is half price on Vertex AI.","openSource":false,"license":"Proprietary","released":"2026-04-22","notes":"Google's first multimodal embedding model, mapping text, images, video, audio, and PDFs into one space. Matryoshka output from 128 to 3072 dimensions, and unlike 001 it re-normalizes truncated vectors automatically. Replaces gemini-embedding-001 as Google's recommended model.","multilingual":true,"url":"https://ai.google.dev/gemini-api/docs/embeddings","mtebAvg":null},{"id":"gemini-embedding-001","name":"gemini-embedding-001","provider":"Google","type":"embedding","dimensions":3072,"maxInputTokens":2048,"pricePer1MTokens":0.15,"pricingNote":"$0.15 per 1M input tokens on Vertex AI ($0.12 batch). The Gemini API pricing page now lists only Gemini Embedding 2. Supports Matryoshka.","openSource":false,"license":"Proprietary","released":"2025-07-14","notes":"Previous Google text embedding, still served with a long runway: earliest shutdown 2028-05-14 on the Gemini API and no sooner than 2028-05-20 on Vertex AI. Strong multilingual. Truncated outputs need manual re-normalization.","multilingual":true,"url":"https://ai.google.dev/gemini-api/docs/embeddings","mtebAvg":68.3},{"id":"text-embedding-005","name":"text-embedding-005","provider":"Google","type":"embedding","dimensions":768,"maxInputTokens":2048,"pricePer1MTokens":0.025,"pricingNote":"$0.000025 per 1,000 input units on Vertex AI ($0.00002 batch). Vertex AI only.","openSource":false,"license":"Proprietary","released":"2024-11-18","notes":"Google budget tier. 768-dim, English-focused, cheapest of the Vertex AI embeddings. Scheduled to retire on Vertex AI on 2027-04-01, so avoid it for new long-lived indexes.","multilingual":false,"url":"https://cloud.google.com/vertex-ai/generative-ai/docs/embeddings","mtebAvg":null},{"id":"embeddinggemma-300m","name":"embeddinggemma-300m","provider":"Google","type":"embedding","dimensions":768,"maxInputTokens":2048,"pricePer1MTokens":null,"pricingNote":"Open weights on Hugging Face (license acceptance required). Free to self-host.","openSource":true,"license":"Gemma Terms of Use","released":"2025-09-04","notes":"Google's 308M-parameter open embedding model for on-device and offline RAG. Matryoshka truncation to 512, 256, or 128 dimensions. MTEB English v2 69.67 and Multilingual v2 61.15 at 768 dimensions (v2 scores).","multilingual":true,"url":"https://ai.google.dev/gemma/docs/embeddinggemma","mtebAvg":69.67},{"id":"mistral-embed","name":"mistral-embed","provider":"Mistral","type":"embedding","dimensions":1024,"maxInputTokens":8000,"pricePer1MTokens":0.1,"pricingNote":"$0.10 per 1M input tokens via la Plateforme.","openSource":false,"license":"Proprietary","released":"2023-12-11","notes":"European data residency option. Solid English/French/German performance. Same pricing tier as Cohere but with longer context. For code, Mistral sells Codestral Embed at $0.15 per 1M tokens.","multilingual":true,"url":"https://docs.mistral.ai/models","mtebAvg":60.7},{"id":"jina-embeddings-v5-text-small","name":"jina-embeddings-v5-text-small","provider":"Jina AI","type":"embedding","dimensions":1024,"maxInputTokens":32768,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens via the Jina API after a one-time 10M free tokens. Also served through Elastic Inference Service.","openSource":true,"license":"CC-BY-NC-4.0","released":"2026-02-18","notes":"677M multilingual model built on Qwen3-0.6B and distilled from Qwen3-Embedding-4B, covering 119+ languages with Matryoshka dimensions from 32 to 1024. MTEB English v2 71.7 and MMTEB 67.7 (v2 scores). Weights are non-commercial; commercial self-hosting goes through Elastic, which now sells Jina models.","multilingual":true,"url":"https://huggingface.co/jinaai/jina-embeddings-v5-text-small","mtebAvg":71.7},{"id":"jina-embeddings-v3","name":"jina-embeddings-v3","provider":"Jina AI","type":"embedding","dimensions":1024,"maxInputTokens":8192,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens via the Jina API. Weights are CC-BY-NC-4.0, so commercial self-hosting needs a license from Elastic.","openSource":true,"license":"CC-BY-NC-4.0","released":"2024-09-18","notes":"Previous-generation open-weights multilingual embedding, superseded by jina-embeddings-v5. Adapter-based: the same model serves retrieval, classification, and separation tasks via task-specific LoRAs.","multilingual":true,"url":"https://jina.ai/embeddings/","mtebAvg":65.5},{"id":"qwen3-embedding-8b","name":"Qwen3-Embedding-8B","provider":"Alibaba Qwen","type":"embedding","dimensions":4096,"maxInputTokens":32000,"pricePer1MTokens":null,"pricingNote":"Open weights. Free to self-host. 0.6B and 4B siblings share the recipe.","openSource":true,"license":"Apache-2.0","released":"2025-06-05","notes":"The open-weight embedding most teams benchmark against. Instruction-aware, 100+ natural and programming languages, output dimensions from 32 to 4096. MTEB English v2 75.22 and MTEB Multilingual 70.58 (v2 scores). The 0.6B version is the most downloaded model in the family for cheap self-hosting.","multilingual":true,"url":"https://huggingface.co/Qwen/Qwen3-Embedding-8B","mtebAvg":75.22},{"id":"nemotron-3-embed-1b","name":"Nemotron-3-Embed-1B","provider":"NVIDIA","type":"embedding","dimensions":2048,"maxInputTokens":32768,"pricePer1MTokens":null,"pricingNote":"Open weights on Hugging Face. Free to self-host.","openSource":true,"license":"OpenMDW-1.1","released":"2026-07-16","notes":"1.14B multilingual retrieval embedding evaluated across 34 languages, with slicing to 1024 or 512 dimensions after re-normalization. NVIDIA reports RTEB 72.38 and MMTEB retrieval 71.04 NDCG@10, against 61.98 and 59.71 for llama-nemotron-embed-vl-1b-v2 on the same tables. Retrieval-focused, so there is no full MTEB average.","multilingual":true,"url":"https://huggingface.co/nvidia/Nemotron-3-Embed-1B-BF16","mtebAvg":null},{"id":"nomic-embed-text-v1.5","name":"nomic-embed-text-v1.5","provider":"Nomic AI","type":"embedding","dimensions":768,"maxInputTokens":8192,"pricePer1MTokens":null,"pricingNote":"Open weights. Free to self-host. Hosted through Nomic Atlas; Nomic does not publish a per-token rate.","openSource":true,"license":"Apache-2.0","released":"2024-02-14","notes":"Open-source English embedding with Matryoshka support (truncate to 256/512/768). Reproducible training data; one of the few fully open embeddings.","multilingual":false,"url":"https://blog.nomic.ai/posts/nomic-embed-matryoshka","mtebAvg":62.3},{"id":"mxbai-embed-large-v1","name":"mxbai-embed-large-v1","provider":"Mixedbread","type":"embedding","dimensions":1024,"maxInputTokens":512,"pricePer1MTokens":null,"pricingNote":"Open weights. Free to self-host. Also available through the paid Mixedbread API.","openSource":true,"license":"Apache-2.0","released":"2024-03-07","notes":"Apache-licensed dense retriever. Strong English performance. Short input limit (512) is the main constraint vs Voyage/Jina.","multilingual":false,"url":"https://www.mixedbread.com/docs/embeddings/overview","mtebAvg":64.7},{"id":"bge-m3","name":"bge-m3","provider":"BAAI","type":"embedding","dimensions":1024,"maxInputTokens":8192,"pricePer1MTokens":null,"pricingNote":"Open weights. Free to self-host. Available through most inference providers (Together, DeepInfra, Replicate).","openSource":true,"license":"MIT","released":"2024-01-30","notes":"Multi-functional, multi-lingual, multi-granularity. Outputs dense + sparse + multi-vector representations from a single model. Strong on long-document multilingual retrieval.","multilingual":true,"url":"https://huggingface.co/BAAI/bge-m3","mtebAvg":64.5},{"id":"rerank-3","name":"rerank-3","provider":"Voyage AI","type":"reranker","dimensions":null,"maxInputTokens":32000,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens (query tokens times documents plus document tokens), the same rate as rerank-2.5. First 200M tokens free for new accounts.","openSource":false,"license":"Proprietary","released":"2026-09-30","notes":"Voyage's recommended reranker, a drop-in upgrade from rerank-2.5 at the same price and the same 32K query-plus-document limit. Vendor figures across 95 retrieval datasets put it 2.72% ahead of Cohere Rerank v4.0 Pro and 3.02% ahead of Qwen3-Reranker-8B, and 3.35% ahead of rerank-2.5 on LongEmbed long-document retrieval. rerank-3-lite is $0.02 per 1M tokens and Voyage says it matches rerank-2.5 quality at 40% of the price.","multilingual":true,"url":"https://docs.voyageai.com/docs/reranker","mtebAvg":null},{"id":"rerank-v4.0-pro","name":"rerank-v4.0-pro","provider":"Cohere","type":"reranker","dimensions":null,"maxInputTokens":32768,"pricePer1MTokens":0,"pricingNote":"$2.50 per 1k searches at last listing. Cohere's public pricing page now shows only Model Vault dedicated-instance rates for rerankers, $5/hour for a Medium instance and $10/hour for a Large, so confirm the per-search rate in the dashboard.","openSource":false,"license":"Proprietary","released":"2025-12-11","notes":"Cohere flagship reranker with a 32K context, four times rerank-v3.5, and support for semi-structured JSON documents. The quality tier for complex enterprise retrieval, though Voyage's rerank-3 release claims a 2.72% lead over it across 95 datasets on Voyage's own harness. A search is one query with up to 100 documents, with longer documents split into chunks.","multilingual":true,"url":"https://docs.cohere.com/docs/rerank","mtebAvg":null},{"id":"rerank-v4.0-fast","name":"rerank-v4.0-fast","provider":"Cohere","type":"reranker","dimensions":null,"maxInputTokens":32768,"pricePer1MTokens":0,"pricingNote":"$2 per 1k searches at last listing. Only the Model Vault dedicated rate ($5/hour for a Medium instance) appears on the current public pricing page.","openSource":false,"license":"Proprietary","released":"2025-12-11","notes":"Lighter Rerank 4 variant for latency-sensitive and high-throughput pipelines, with the same 32K context as Pro.","multilingual":true,"url":"https://docs.cohere.com/docs/rerank","mtebAvg":null},{"id":"rerank-v3.5","name":"rerank-v3.5","provider":"Cohere","type":"reranker","dimensions":null,"maxInputTokens":4096,"pricePer1MTokens":0,"pricingNote":"$2 per 1k searches at last listing. Cohere's public pricing page now lists only Rerank 4 per-search rates.","openSource":false,"license":"Proprietary","released":"2024-12-03","notes":"Previous Cohere reranker, still served on the Cohere API. Rerank 4 quadruples the context window. Azure retired its hosted copy on 2026-05-14.","multilingual":true,"url":"https://docs.cohere.com/docs/rerank","mtebAvg":null},{"id":"rerank-2.5","name":"rerank-2.5","provider":"Voyage AI","type":"reranker","dimensions":null,"maxInputTokens":32000,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens (query tokens times documents plus document tokens).","openSource":false,"license":"Proprietary","released":"2025-08-11","notes":"Moved to Voyage's older-models list on 2026-09-30 when rerank-3 shipped at the same price with the same 32K limit, so there is no cost reason to start here. Still served, with instruction-following and multilingual support. rerank-2.5-lite is $0.02 per 1M tokens and is likewise superseded by rerank-3-lite.","multilingual":true,"url":"https://docs.voyageai.com/docs/reranker","mtebAvg":null},{"id":"rerank-2","name":"rerank-2","provider":"Voyage AI","type":"reranker","dimensions":null,"maxInputTokens":16000,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens (query + documents combined). Listed under older models.","openSource":false,"license":"Proprietary","released":"2024-08-09","notes":"Two generations behind now: superseded by rerank-2.5 at the same price with double the context, and then by rerank-3 in September 2026. rerank-2-lite caps at 8,000 tokens.","multilingual":true,"url":"https://docs.voyageai.com/docs/reranker","mtebAvg":null},{"id":"jina-reranker-v3.5","name":"jina-reranker-v3.5","provider":"Jina AI","type":"reranker","dimensions":null,"maxInputTokens":131072,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens via the Jina API.","openSource":true,"license":"CC-BY-NC-4.0","released":"2026-07-27","notes":"Listwise reranker on Qwen3-0.6B that scores a whole candidate list jointly within a 131K context. Jina reports it beats a 4B Qwen reranker on BEIR at about a seventh of the size. Non-commercial weights.","multilingual":true,"url":"https://huggingface.co/jinaai/jina-reranker-v3.5","mtebAvg":null},{"id":"jina-reranker-v2","name":"jina-reranker-v2-base-multilingual","provider":"Jina AI","type":"reranker","dimensions":null,"maxInputTokens":1024,"pricePer1MTokens":0.05,"pricingNote":"$0.05 per 1M tokens via Jina API. Weights are CC-BY-NC-4.0.","openSource":true,"license":"CC-BY-NC-4.0","released":"2024-07-04","notes":"Older open-weights reranker, 100+ languages. Superseded by jina-reranker-v3 and v3.5 at the same API price with far longer context.","multilingual":true,"url":"https://jina.ai/reranker/","mtebAvg":null},{"id":"qwen3-reranker-4b","name":"Qwen3-Reranker-4B","provider":"Alibaba Qwen","type":"reranker","dimensions":null,"maxInputTokens":32000,"pricePer1MTokens":null,"pricingNote":"Open weights. Free to self-host. 0.6B and 8B sizes also available.","openSource":true,"license":"Apache-2.0","released":"2025-06-05","notes":"The most downloaded open reranker on Hugging Face and the commercially clean alternative to the non-commercial Jina weights. Instruction-aware, 100+ languages, 32K context. Pairs with Qwen3-Embedding.","multilingual":true,"url":"https://huggingface.co/Qwen/Qwen3-Reranker-4B","mtebAvg":null}]}