NAVIN RELEASES
A home for the models we build and release.
| # | Model / system | Type | Details | Release date | Version | Size |
|---|---|---|---|---|---|---|
| 01 | Dossier-1BPreparing | Document | Fast document OCR, handwriting, layout analysis, tables, forms, receipts and key-value extraction. | — | — | 1B |
| 02 | Dossier-2BPreparing | Document | Compact document understanding with structured extraction, classification, document QA and field-level validation. | — | — | 2B |
| 03 | Dossier-4BPreparing | Agentic Document | Flagship Agentic Document Intelligence model. OCR, document reasoning, evidence grounding, verification, risk detection, tool use and bounded repair. | — | — | 4B |
| 04 | Dossier-9BPreparing | Document Reasoning | Advanced multi-page and multi-document reasoning for complex contracts, financial documents, compliance workflows and high-quality verification. | — | — | 9B |
| 05 | Sandora-3BPreparing | Agent | Lightweight agent model for tool calling, routing, delegation, local automation and basic computer-use tasks. | — | — | 3B |
| 06 | Sandora-9BPreparing | Adaptive Agent | Flagship Adaptive Multimodal Agent. Planning, computer use, multi-agent orchestration, tool use, memory, skill learning, self-reflection and user adaptation. | — | — | 9B |
| 07 | Sandora-27BPreparing | Agent Reasoning | High-capability agent model for long-horizon planning, complex orchestration, reasoning, trajectory generation, evaluation and teacher-model workloads. | — | — | 27B |
| 08 | Moyi-T2.5-160MPreparing | Speech | Micro / embedded streaming speech model optimized for highly constrained edge devices and offline execution. | — | — | 160M |
| 09 | Moyi-T2.5-350MPreparing | Speech | Ultra-edge multilingual speech intelligence for low-memory mobile and embedded hardware. | — | — | 350M |
| 10 | Moyi-T2.5-600MPreparing | Edge Speech | Flagship Edge Speech Intelligence model. Real-time multilingual speech translation, streaming inference, context awareness and on-device execution. | — | — | 600M |
| 11 | Moyi-T2.5-1.7BPreparing | Speech | Quality-oriented speech model for workstation, higher-end edge devices and teacher / distillation workloads. | — | — | 1.7B |