Sarvam-30B
Active Parameters
32B
Context Length
128K
Modality
Text
Architecture
Mixture of Experts (MoE)
License
Apache 2.0
Release Date
6 Mar 2026
Knowledge Cutoff
-
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
4x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
128,000 tokens
Consumer
4x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
No evaluation benchmarks for Sarvam-30B available.
Rankings
Overall Rank
-
Coding Rank
-
About Sarvam-30B
Sarvam-30B is an advanced Mixture-of-Experts (MoE) model with 32B total parameters and 2.4B active parameters, designed for practical deployment in resource-constrained environments. Released March 6, 2026 under Apache 2.0 license. Uses 19 layers with 128 experts, top-6 routing, grouped KV attention (4 heads), and extremely high rope_theta (8e6) for long-context stability. Delivers state-of-the-art performance across 22 Indian languages with strong reasoning, reliable coding ability, and best-in-class conversational quality. Optimized for multilingual voice calls with tool calling capabilities, throughput, and memory efficiency.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
64
Key-Value Heads
4
Attention Head Dimension
64
Position Embedding
ROPE
RoPE Theta
8,000,000
Sliding Window Attention
No
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
4,096
Number of Layers
19
FFN Intermediate Size (Dense)
1,024
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
262,144
Mixture of Experts
Total Expert Parameters
2.4B
Number of Experts
128
Active Experts
6
Shared Experts
1
FFN Intermediate Size (per Expert)
1,024
Dense Layers Before MoE
1
About Sarvam
Sarvam AI's sovereign foundation models built for India's languages, culture, and context. Released in March 2026, these advanced Mixture-of-Experts (MoE) models offer state-of-the-art performance across 22 Indian languages while maintaining competitive results on global benchmarks. Designed with focus on reasoning, coding, multilingual capabilities, and agentic tasks. Open-sourced under Apache 2.0 license, optimized for practical deployment from resource-constrained environments to high-performance applications.