BGE Models by Beijing Academy of Artificial Intelligence (BAAI)
About
BGE (BAAI General Embedding) is the Beijing Academy of Artificial Intelligence's open-source family of embedding and reranking models, developed under the FlagEmbedding project. BGE models are among the most-downloaded embedding models on Hugging Face and consistently rank near the top of the MTEB leaderboard. The family spans English-only and multilingual embedding models, cross-encoder rerankers, and large LLM-based embedding models such as BGE-Multilingual-Gemma2. NVIDIA NIM also serves BGE models (including BGE-M3) in its hosted retrieval API catalog.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
Use when the workload needs embedding, 4k context, and 9B parameters.
Use when the workload needs 8k context and 568M parameters.
Use when the workload needs embedding, 8k context, and 568M parameters.
Use when the workload needs embedding, 512 context, and 335M parameters.
Use when the workload needs ranking, 512 context, and 560M parameters.
Use when the workload needs embedding, 512 context, and 109M parameters.
Use when the workload needs embedding, 512 context, and 33.4M parameters.
Use when the workload needs ranking and 512 context.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| BGE Multilingual Gemma2 | Use when the workload needs embedding, 4k context, and 9B parameters. | 2024-06 | embedding4k context9B parameters | Current |
| BGE Reranker V2 M3 | Use when the workload needs 8k context and 568M parameters. | 2024-03 | 8k context568M parameters | Current |
| BGE M3 | Use when the workload needs embedding, 8k context, and 568M parameters. | 2024-01 | embedding8k context568M parameters | Current |
| BGE Large EN v1.5 | Use when the workload needs embedding, 512 context, and 335M parameters. | 2023-09 | embedding512 context335M parameters | Current |
| BGE Reranker Large | Use when the workload needs ranking, 512 context, and 560M parameters. | 2023-09 | ranking512 context560M parameters | Current |
| BGE Base EN v1.5 | Use when the workload needs embedding, 512 context, and 109M parameters. | 2023-09 | embedding512 context109M parameters | Current |
| BGE Small EN v1.5 | Use when the workload needs embedding, 512 context, and 33.4M parameters. | 2023-09 | embedding512 context33.4M parameters | Current |
| BGE Reranker Base | Use when the workload needs ranking and 512 context. | 2023-09 | ranking512 context | Current |
Release Timeline
4 release groupsSpecifications(8 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| BGE Multilingual Gemma2 | 2024-06 | 4k | 9B |
| BGE Reranker V2 M3 | 2024-03 | 8k | 568M |
| BGE M3 | 2024-01 | 8k | 568M |
| BGE Large EN v1.5 | 2023-09 | 512 | 335M |
| BGE Reranker Large | 2023-09 | 512 | 560M |
| BGE Base EN v1.5 | 2023-09 | 512 | 109M |
| BGE Small EN v1.5 | 2023-09 | 512 | 33.4M |
| BGE Reranker Base | 2023-09 | 512 | — |
Available From(2 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| BGE M3 | Novita AI | $0.01 | — | Serverless |
| BGE Reranker V2 M3 | Novita AI | $0.01 | — | Serverless |
Frequently Asked Questions
- What is BGE used for?
- BGE is used for embedding, ranking, and coding. The family description and listed model capabilities point to those workloads as the best fit.
- How does BGE compare to Aquila 2?
- BGE by Beijing Academy of Artificial Intelligence (BAAI) is strongest where you need embedding, while Aquila 2 by Beijing Academy of Artificial Intelligence (BAAI) is the closest related family to check for coding. BGE has 8 listed variants and reaches up to 8k context, while Aquila 2 reaches up to 16k context, so compare the specs and pricing tables before choosing a production model.
- Which BGE model should I use?
- For the lowest listed input price, start with BGE M3 through Novita AI at $0.01/1M input tokens. For the most capable/latest local choice, evaluate BGE Reranker V2 M3 with 8k context.
Models(8)
BGE Multilingual Gemma2
BGE Reranker V2 M3
BGE M3
BGE Large EN v1.5
BGE Reranker Large
BGE Base EN v1.5
BGE Small EN v1.5
BGE Reranker Base