Sorika Models
Frontier neural foundation models engineered by Darsh Yadav for lightweight speech synthesis and spatial document intelligence.
Active Foundation Engines
Swen-1.1-Instruct
Hybrid Conv-Attention (1.2B)High-efficiency compact language model scoring 66.0% on HumanEval, 54.0% on GSM8K, with native tool use and 128k context.
Swen-1-Math
Chain-of-Thought (350M)Ultra-compact mathematical specialist scoring 85.0% on MultiArith and 35.0% on MATH-500, running entirely in CPU cache.
Swen-1.1-Thinking
Self-Reflection (1.2B)Metacognitive reasoning engine for self-reflection and backward induction in game theory and combinatorial puzzles at 95 tok/s.
Suzune S1 Voice
Non-Autoregressive TTSUltra-lightweight 80M non-autoregressive voice foundation model with sub-20ms latency and 24kHz studio HD audio.
Kaori K1 Spatial OCR
Spatial Vision120M spatial layout transformer for zero-shot extraction across multi-column PDFs, scientific formulas, and Indic manuscripts.
Upcoming Architectures in Training
Next-generation models currently being researched, trained, and benchmarked by Darsh Yadav at Sorika Labs.
Haruka Discovery
High-throughput neural discovery engine for multimodal real-time horizon retrieval across dense vector spaces.
Swen Latent
Continuous spatial latent embeddings for multi-hop graph discovery and zero-latency semantic vector indexing.
Sorika Logic Proofs
Deterministic multi-step reasoning foundation engine engineered for verifiable symbolic proofs and mathematical verification.