small-language-models
(7)
Mark McQuade Started Small. Then He Bet $20 Million on Open-Weight AI
Arcee AI began by making other people’s models more useful. Three years later, McQuade and a compact team trained a 400-billion-parameter model of their own - and turned a career spent translating technology into a wager on ownership.
Interfaze
Interfaze is a San Francisco AI startup building a model purpose-built for deterministic developer tasks - OCR, web scraping, speech-to-text, object detection and classification - where consistency and verifiable output matter more than open-ended chat. Its hybrid architecture fuses specialized neural networks (CNN/DNN) directly into a transformer decoder, and the model ships as an OpenAI-compatible API with confidence scores, bounding boxes and metadata so developers can build rule-based systems on top. Founded by Yoeven Khemlani and Harsha Vardhan Khurdula, Interfaze is part of Y Combinator's 2026 batch and grew out of the team's earlier developer-API company, JigsawStack.
DXwand
DXwand is a Dubai- and Cairo-based enterprise AI company whose ORXTRA platform lets large organisations build and deploy generative AI agents on top of their own documents, databases and workflows. Founded in 2018 by two former Microsoft executives, Ahmed Mahmoud and Mahmoud Gomaa, the company built its edge on two hard problems most Western vendors skipped: understanding spoken Arabic across its many dialects, and running useful AI cheaply enough that a bank or a ministry can afford to put it in front of millions of people. Its knowledge-mining and RAG engine removes the manual data-labelling step that normally stalls enterprise AI projects, and its agents ship as pre-built modules for customer service, procurement and HR, or as custom builds through a no-code Agent Builder that supports 20+ language models and 80+ data sources. DXwand serves financial services, government, healthcare, telecom, education, media and retail clients across MENA, has raised roughly $6.9 million to date including a $4 million Series A led by Shorooq Partners and Algebra Ventures, and says it reached about $5 million in annual recurring revenue while remaining profitable.
Liquid AI
Liquid AI is an MIT CSAIL spinout building Liquid Foundation Models (LFMs), a family of general-purpose AI models engineered to run fast and privately on-device - phones, laptops, wearables, and cars - instead of relying on the cloud. Rooted in the founders' 'liquid neural network' research inspired by the nervous system of a roundworm, the company pairs a first-principles model architecture with LEAP, a developer platform that deploys models into mobile apps in a few lines of code, and Apollo, a fully local AI chat app. It raised a $250M Series A led by AMD in December 2024 at a roughly $2.35B valuation.
Arcee AI
Arcee AI is a San Francisco-based open intelligence lab that builds small, efficient, US-made language models enterprises can own, customize and run inside their own infrastructure. Founded in 2023 by former Hugging Face and Roboflow operators, the company pioneered practical techniques like model merging (via its acquisition of mergekit) and Spectrum training, and has expanded from small language models into its own open-weight Arcee Foundation Model family, the Trinity mixture-of-experts models, and agentic products such as Arcee Conductor and Arcee Orchestra. Its pitch: smaller, domain-adapted models that rival frontier systems at a fraction of the cost while keeping data private.
Fawad Butt
Fawad Butt is the co-founder and CEO of Penguin AI, a Palo Alto-based healthcare AI platform company that raised $29.7M in Series A funding in September 2025. A former Chief Data Officer at Kaiser Permanente, UnitedHealthcare, and Optum, Butt spent years on the buyer side spending hundreds of millions on generic tech before deciding to build the solution healthcare actually needed. Penguin AI targets the $1 trillion annual administrative burden in US healthcare with purpose-built agentic AI, small language models, and digital workers that compress prior authorization reviews from 30 minutes to under two, positioning itself as 'the Epic of the healthcare back office.'
Predibase
Predibase was a San Francisco-based AI infrastructure company (founded 2020, acquired by Rubrik in June 2025) that pioneered efficient LLM fine-tuning and serving at scale. Built by the creators of Uber AI's Ludwig and Horovod frameworks, Predibase made it easy for enterprises to fine-tune and deploy open-source LLMs using LoRA adapters — often outperforming GPT-4 on domain-specific tasks for under $8 of compute. Its open-source LoRAX inference server enabled serving thousands of fine-tuned models from a single GPU, dramatically cutting costs. After raising $28M from Greylock and Felicis, Predibase was acquired by cybersecurity firm Rubrik for over $100M to accelerate agentic AI adoption.