BGE M3
BAAI · dense · 0.568B parameters · 8,192 context
Parameters
0.568B
Context Window
8K tokens
Architecture
Dense
Best GPU
B200 SXM
Cheapest API
$0.01/M
Intelligence Brief
BGE M3 is a 0.568B parameter DENSE model from BAAI, featuring Multi-Head Attention (MHA) with 24 layers and 1,024 hidden dimensions. With a 8,192 token context window, it supports multilingual. The most cost-effective API deployment is via together at $0.01/M output tokens. For self-hosted inference, B200 SXM delivers optimal throughput at $4261/month.
Architecture Details
Memory Requirements
BF16 Weights
1.1 GB
FP8 Weights
0.6 GB
INT4 Weights
0.3 GB
GPU Compatibility Matrix
BGE M3 is compatible with 100% of GPU configurations across 41 GPUs at 3 precision levels.
GPU Recommendations
FP8 · 1 GPU · tensorrt-llm
83/100
score
Throughput
3.5K tok/s
Latency (ITL)
0.3ms
Est. TTFT
0ms
Cost/Month
$4261
Cost/M Tokens
$0.46
FP8 · 1 GPU · tensorrt-llm
83/100
score
Throughput
3.5K tok/s
Latency (ITL)
0.3ms
Est. TTFT
0ms
Cost/Month
$4271
Cost/M Tokens
$0.46
FP8 · 1 GPU · tensorrt-llm
83/100
score
Throughput
3.5K tok/s
Latency (ITL)
0.3ms
Est. TTFT
0ms
Cost/Month
$6169
Cost/M Tokens
$0.67
Deployment Options
API Deployment
together
$0.01/M
output tokens
Single GPU
B200 SXM
$4261/mo
Min VRAM: 1 GB
Multi-GPU
B200 SXM
3.5K tok/s
Best available config
API Pricing Comparison
| Provider | Input $/M | Output $/M | Badges |
|---|---|---|---|
| together | $0.01 | $0.01 | Cheapest |
Cost Analysis
| Provider | Input $/M | Output $/M | ~Monthly Cost |
|---|---|---|---|
| togetherBest Value | $0.01 | $0.01 | $0 |
Cost per 1,000 Requests
Short (500 tok)
$0.01
via together
Medium (2K tok)
$0.02
via together
Long (8K tok)
$0.08
via together
Performance Estimates
Throughput by GPU
VRAM Breakdown (B200 SXM, FP8)
Precision Impact
bf16
1.1 GB
weights/GPU
fp8
0.6 GB
weights/GPU
~3.5K tok/s
Capabilities
Features
Supported Frameworks
Supported Precisions
Where to Deploy BGE M3
Self-Hosted Infrastructure
Similar Models
BGE Large EN v1.5
0.335B params · dense
Quality: 50
from $0.01/M
Jina Embeddings v3
0.57B params · dense
Quality: 50
from $0.01/M
Multilingual E5 Large
0.56B params · dense
Quality: 50
InfoXLM Large
0.55B params · dense
Quality: 50
Parakeet CTC 0.6B
0.6B params · dense
Quality: 50
from $0.03/M
Frequently Asked Questions
How much VRAM does BGE M3 need for inference?
BGE M3 requires approximately 1.1 GB of VRAM at BF16 precision, 0.6 GB at FP8, or 0.3 GB at INT4 quantization. Additional VRAM is needed for KV-cache (98304 bytes per token) and activations (~0.10 GB).
What is the best GPU for BGE M3?
The top recommended GPU for BGE M3 is the B200 SXM using FP8 precision. It achieves approximately 3.5K tokens/sec at an estimated cost of $4261/month ($0.46/M tokens). Score: 83/100.
How much does BGE M3 inference cost?
BGE M3 API inference starts from $0.01/M input tokens and $0.01/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.