Empirical Database

AI Models Architecture & Benchmark Index

A structured catalog analyzing frontier foundation models, sovereign Indic architectures, and edge-quantized variants across parameter counts, context limits, and reasoning capability.

Sym	Model Name	Maker / Organization	Architecture Family	Parameters	Context	Eval Score	Throughput
Cs	Claude Sonnet 5.5 Optimal balance of frontier reasoning, coding, and cost efficiency.	Anthropic	Flagship Dense	Undisclosed	200k	93.8%	85 tps
Co	Claude Opus 5.5 Deep complex reasoning, research synthesis, and scientific proofs.	Anthropic	Flagship Dense	Undisclosed	200k	95.2%	45 tps
Bg	BharatGen Param2 Sovereign Indian Foundation Model natively supporting all 22 scheduled languages.	IIT Bombay / IndiaAI	Sovereign Indic	17B MoE	128k	84.5%	120 tps
Sv	Sarvam-M High-efficiency Indic language foundation model for public administration.	Sarvam AI	Sovereign Indic	24B Dense	64k	82.1%	110 tps
G3	Gemini 3.1 Pro Massive 2M token context window with native multimodal video comprehension.	Google DeepMind	Multimodal Native	Undisclosed	2M	92.4%	90 tps
L3	Llama 3.3 70B State-of-the-art open-weight baseline for enterprise self-hosting.	Meta AI	Open-weight	70B Dense	128k	88.6%	78 tps
D3	DeepSeek V3 Multi-head Latent Attention with ultra-low training compute overhead.	DeepSeek	Mixture-of-Experts	671B (37B active)	128k	91.5%	65 tps
Tq	TurboQuant-Llama3 Compressed 70B parameter footprint running in 19.8GB memory for consumer NPUs.	BluePatterns AI	Efficient / Edge	70B (INT3 / FP8)	32k	87.8%	138 tps
Sk	Soket 120B Homegrown sovereign model designed for national defense and critical healthcare.	Soket AI	Sovereign Indic	120B Dense	64k	83.2%	55 tps
Cc	Cursor Composer 2.0 Multi-file agentic code generation and diff maintenance across repositories.	Anysphere / Cursor	Code Specialist	Agent Swarm	256k	94.1%	95 tps