Marco O1
Alibaba · dense · 7.6B parameters · 65,536 context
Parameters
7.6B
Context Window
64K tokens
Architecture
Dense
Best GPU
A10G
Intelligence Brief
Marco O1 is a 7.6B parameter DENSE model from Alibaba, featuring Grouped Query Attention (GQA) with 28 layers and 3,584 hidden dimensions. With a 65,536 token context window, it supports math, multilingual, reasoning. For self-hosted inference, A10G delivers optimal throughput at $285/month.
Architecture Details
Memory Requirements
BF16 Weights
15.2 GB
FP8 Weights
7.6 GB
INT4 Weights
3.8 GB
GPU Compatibility Matrix
Marco O1 is compatible with 95% of GPU configurations across 41 GPUs at 3 precision levels.
GPU Recommendations
BF16 · 1 GPU · vllm
100/100
score
Throughput
213.1 tok/s
Latency (ITL)
4.7ms
Est. TTFT
1ms
Cost/Month
$285
Cost/M Tokens
$0.51
BF16 · 1 GPU · vllm
100/100
score
Throughput
331.4 tok/s
Latency (ITL)
3.0ms
Est. TTFT
1ms
Cost/Month
$332
Cost/M Tokens
$0.38
BF16 · 1 GPU · vllm
100/100
score
Throughput
358.1 tok/s
Latency (ITL)
2.8ms
Est. TTFT
0ms
Cost/Month
$370
Cost/M Tokens
$0.39
Deployment Options
API Deployment
No API pricing available
Single GPU
A10G
$285/mo
Min VRAM: 8 GB
Multi-GPU
A4000 x2
248.0 tok/s
TP· $323/mo
API Pricing Comparison
No API pricing data available for this model.
Performance Estimates
Throughput by GPU
VRAM Breakdown (A10G, BF16)
Precision Impact
bf16
15.2 GB
weights/GPU
~213.1 tok/s
fp8
7.6 GB
weights/GPU
int4
3.8 GB
weights/GPU
Capabilities
Features
Supported Frameworks
Supported Precisions
Where to Deploy Marco O1
Self-Hosted Infrastructure
Similar Models
GTE Qwen2 7B
7.6B params · dense
Quality: 50
from $0.02/M
Qwen 2 Audio 7B
7.6B params · dense
Quality: 50
Qwen 2.5 7B
7.6B params · dense
Quality: 70
from $0.20/M
Qwen 2.5 Coder 7B
7.6B params · dense
Quality: 50
from $0.20/M
Qwen 2.5 Math 7B
7.6B params · dense
Quality: 50
from $0.20/M
Frequently Asked Questions
How much VRAM does Marco O1 need for inference?
Marco O1 requires approximately 15.2 GB of VRAM at BF16 precision, 7.6 GB at FP8, or 3.8 GB at INT4 quantization. Additional VRAM is needed for KV-cache (57344 bytes per token) and activations (~0.80 GB).
What is the best GPU for Marco O1?
The top recommended GPU for Marco O1 is the A10G using BF16 precision. It achieves approximately 213.1 tokens/sec at an estimated cost of $285/month ($0.51/M tokens). Score: 100/100.
How much does Marco O1 inference cost?
Marco O1 inference costs vary by provider and GPU setup. Use our calculator for detailed cost estimates across all providers.