bge-m3
Official · BF16 · safetensors
Runtime certified: Transformers Sentence-transformers
| Precision / quantization | Format | Bits | Publisher trust | Deployment state |
|---|---|---|---|---|
| BF16 | safetensors | 16 | Official | Runtime certified |
BGE · Embedding · BGE-M3
BGE-M3: Specs, GPU requirements, hosting & deployment
BGE-M3 is a current Embedding release in the BGE family, certified from its immutable official source revision.
Total parameters568M
Active parametersN/A / Not directly comparable
Context window8,192 tokens
Release statusCurrent · Current
ArchitectureMultilingual embedding transformer
ModalitiesText
LicenseMIT License
Deployment stateDeployable with certified configuration
Commercial stateCommercial Use Allowed
Release date2024-01-27
Architecture, strengths and use cases
BGE-M3 is an official self-hostable BAAI Embedding and reranking release. It uses Multilingual embedding transformer with 0.568B. Exact checkpoint, license, runtime, and hardware evidence is tied to the verified upstream revision.
BGE-M3 uses Multilingual embedding transformer with 0.568B. Pipeline and configuration fields are reported only where the official source publishes them.
Verified Multilingual embedding transformer architecture
Official checkpoint and pipeline evidence
Official BF16, Unknown checkpoint coverage
Self-hosted Embedding and reranking workloads
Controlled model evaluation
Production deployment planning
Certified checkpoints and quantizations
Only exact, source-certified variants attached to this release are shown.
Recommended hardware
Use only the listed runtime configurations with the exact checkpoint precision. Hardware sizing considers resident pipeline memory, runtime overhead, input and output dimensions, batch size, and safety headroom.
Estimated weight memory1.09 GB
Runtime overhead4.92 GB
Total with safety margin7.45 GB
Recommended best value
1× NVIDIA A100 SXM 80GB
80 GB total VRAM · nvlink
72.55 GB headroom · Available
USD 915 / monthLowest-cost viable
1× NVIDIA L4 24GB
24 GB total VRAM · pcie
16.55 GB headroom · Available
USD 250 / monthHighest performance
1× NVIDIA H200 SXM 141GB
141 GB total VRAM · nvswitch
133.55 GB headroom · Limited capacity
USD 2,345 / monthLicense and self-hosting considerations
The exact pipeline and selected precision must fit on the chosen GPUs. Requirements vary with resolution, duration, batch size, modalities, and runtime; language-model KV-cache rules are not applied.
BGE-M3 is published under MIT License. The release-scoped official license link and verification date are retained with the catalogue record.
Commercial Use Allowed. The official license permits commercial use; its obligations still apply.
Alternatives and sibling variants
bge-m3, IndustryCorpus2_DataRater, IndustryCorpus2_Classifier
Release timeline
baai/bge/bge-m3 @ 4796263f4dddb70099a9147ad014ef6972a289db9d64a2bcf62891798b71dff2
BAAI/bge-m3 @ 2024-09-14T00:26:03Z
Frequently asked questions
How much VRAM does BGE-M3 require?
The estimate uses the certified checkpoint, precision, runtime, workload dimensions, batch size, and domain-specific overhead.
Which runtimes support BGE-M3?
Certified upstream runtime evidence is available for transformers, sentence-transformers.
Can BGE-M3 be used commercially?
The stored commercial classification is Commercial Use Allowed. Consult the official MIT License terms for the final obligations.
Official references
Last verified: 2026-08-26