Active Parameters
80B
Context Length
262K
Modality
Text
Architecture
Mixture of Experts
KV Cache Rate
96 KB/tok
License
Apache 2.0
Release Date
12 Sept 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Self-hosted only
Precision
Context
Workload Memory: 157.85 GB
FP16 · 8K context
Base Weights: 154.60 GB
KV Cache: 0.75 GB
Runtime & Graphs: 2.50 GB
8x NVIDIA H100 80GB
157.8 / 576.0 GB usable · TP 8
8x RTX 4090 24GB
157.8 / 172.8 GB usable · TP 8
Determine requirements for your specific workload:
Open in VRAM Calculator
Checkpoint format
FP16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 5.16.1
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native FP16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native FP16 | FlashAttention-2 |
Additional required packages
Install alongside standard runtime libraries:
pip install fla
No evaluation benchmarks for AliceAI Foundation 80B A3B Base available.
Overall Rank
-
Coding Rank
-
AliceAI Foundation 80B A3B Base is a sparse mixture-of-experts foundation model developed by Yandex, routing tokens through 3B active parameters out of an 80B total parameter pool. It provides strong multilingual and reasoning abilities with high computational efficiency during inference.
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
16
Key-Value Heads
2
Attention Head Dimension
256
Position Embedding
ROPE
RoPE Theta
1,000,000
Linear Attention
Yes
Linear Attention Ratio
75.0%
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Base Precision
FP16
Hidden Dimension Size
2,048
Number of Layers
48
Multi-Token Prediction Heads
1
Tokenizer
Vocabulary Size
129,024
Mixture of Experts
Total Expert Parameters
3.0B
Number of Experts
512
Active Experts
10
Shared Experts
1
FFN Intermediate Size (per Expert)
512
Dense Layers Before MoE
0
The AliceAI model family developed by Yandex.