PioneeringresearchonthepathtoAGI.
We believe our research will eventually lead to artificial general intelligence, a system that can solve human-level problems. Our mission is to ensure that AGI benefits all of humanity.
SAGE
SAGE series models are efficient, reasoning-focused AI systems designed to break down and solve complex problems through agentic planning and iterative distillation.
SAGE 2.4 Actus-bio
A clinical reasoning model built to address the lack of AI support for frontline health workers operating in Nepali, Maithili, Bhojpuri and other Indic languages. The model employs Recursive Confidence Calibration to explicitly navigate the ambiguity of differential diagnosis.
Introducing SAGE-OSS-40B
Open-sourcing SAGE-OSS-40B, a 40B LoopCoder mixture-of-experts research model released under Apache 2.0.
Introducing SAGE 2.4 Actus
The evolution of SAGE-32B into a fully realized agentic reasoning platform, leading our research trajectory.
Introducing Rune
Rune is a local-first CLI coding assistant powered by SAGE models. Code with AI entirely on your own machine—private, fast, and fully offline.
SAGE-32B
A 32 billion parameter language model that focuses on agentic reasoning and long range planning tasks. SAGE-32B achieves higher success rates in multi-tool usage scenarios compared to similarly sized baseline models.
SAGE on Ollama
We have released Ollama ports for SAGE-v1 models. Run our efficient reasoning models locally on your device.
Introducing SAGE 2.5 Celer
Announcing the first reasoning models in the SAGE series, combining extreme efficiency with advanced problem-solving capabilities. Featuring SAGE Celer Low 2.5 (3B), SAGE Celer Mid 2.5 (8B), and SAGE Celer High 2.5 (14B).
Introducing SAGE 1.5 Celer
An early internal iteration of our IDA-based reasoning models, paving the way for advanced efficiency. While less advanced than the current specs of SAGE, it demonstrated foundational reasoning capabilities.
Thinking About Thinking
We introduce inverse reasoning, a novel paradigm enabling LLMs to decompose and explain their own reasoning chains post-hoc. Our approach, used in SAGE-nano, a 4-billion-parameter reasoning model, employs a metacognitive structure that reflects back via attention processes to identify major decision points and generate explanations of reasoning choices. While typical CoT approaches are directed towards forward reasoning generation, inverse reasoning provides insight into why specific reasoning chains were selected over others.
SAGE Nano 1.x
We introduce SAGE-nano, a compact reasoning model that achieves state-of-the-art performance on complex reasoning tasks, aided with what we call inverse reasoning. SAGE-nano utilizes bidirectional chain-of-thought processing to enhance reasoning capabilities in resource-constrained environments.
Introducing SAGE Nano 1.x
SAGE is our most efficient large language model, delivering exceptional reasoning capabilities with breakthrough efficiency, even on CPU hardware.
VORA
VORA models redefine voice synthesis, combining hyperrealistic audio generation with unprecedented efficiency and emotional intelligence for natural human-computer interaction.
VORA v1 & L1 System Card
VORA v1 and VORA-L1 combine state-of-the-art voice synthesis with unprecedented efficiency—supporting 40+ languages, real-time emotional adaptation, voice cloning, and seamless multilingual transitions. VORA v1 excels at hyperrealistic speech generation for cloud environments, while VORA-L1 delivers exceptional voice quality on resource-constrained edge devices with just 4MB memory footprint. Both models leverage hierarchical attention mechanisms and acoustic quantization to produce natural-sounding speech that captures the full spectrum of human vocal expression.
VORA V1 and L1
We achieved a breakthrough in AI model efficiency and performance with VORA V1 and L1, setting new standards for Voice Synthesis capabilities.
Introducing VORA-L1
Lightweight, fast, and powerful—VORA-L1 is our first model to be trained with a new architecture that allows for faster inference and lower latency
Introducing VORA v1
Our smartest and most capable models to date with full tool access
Foundational Research
Pioneering efforts across diverse modalities, including computer vision, multimodal intelligence, and fundamental architecture design.
Arva OCR
A 3B-parameter vision-language model establishing new benchmarks for complex layout parsing, specifically tuned for both printed and handwritten Nepali scripts.
Learning to Read from Fragments
We propose GLUE (Generative Ligature & Union Engine), a spatial and procedural optimization framework that employs Poisson boundary blending to enforce gradient continuity across character junctions, mathematically eliminating traditional Devanagari junction seam artifacts.
Introducing v1
Our breakthrough TTS model combines exceptional voice quality with unmatched efficiency. Capable of generating expressive, emotionally nuanced speech in under 100ms, enabling seamless integration into interactive applications.
Build your own AI future
Build, customize, and deploy AI solutions with complete control.