The archive · AI & Models · Strategic decision · 2025–2026
Agnes AI bets a Singapore-built AI app can win the long tail; 3M users, $10M Series A
Singapore AI workspace Agnes launched July 2025, hit 3M users and 200K DAU by December, then a $10M Series A at a $100M+ valuation.
Agnes AI
What the business is
Agnes AI is an all-in-one AI assistant developed and trained in Singapore — search, deep research, slide generation, design, sheets and group chat — running on proprietary models built from scratch.
Starting capital：$10M Series A closed February 13, 2026, led by LOOK FORWARD VCC.
How it started
Bruce Yang — a Raffles Institution alumnus who studied at UC Berkeley, worked at Microsoft and LinkedIn, and built a Silicon Valley product with millions of downloads — came back to Singapore for an AI PhD at NUS. ChatGPT's arrival convinced him most AI companies would fight over paying users in rich markets and ignore the rest of the world, so he assembled a team from NUS, NTU, MIT, Stanford and Berkeley under parent company SAPIENS to build an AI platform for that long tail.
What happened
Agnes launched on Product Hunt on July 4, 2025 and crossed 2 million registered users with about 150,000 daily actives within four months, roughly half from Southeast Asia (TNGlobal). By December 2025 it reported 3 million users and 200,000 DAU, top-10 Google Play productivity rankings in the Philippines, Vietnam and Indonesia, and a proprietary 7B model (Agnes-R1) claimed SOTA on reasoning benchmarks; it also partnered with Agora to add real-time multi-agent group chat.
How it ended up
Still scaling: in February 2026 it closed a $10M Series A led by LOOK FORWARD VCC, and by April 2026 the company said its broader ecosystem served nearly 10 million users and was approaching $20M ARR, with developer API services planned.
Background
Agnes AI is a Singapore-based all-in-one AI assistant — search, deep research, slide generation, design, spreadsheets and group chat in a single app — built and trained in Singapore on proprietary models. Founder Bruce Yang, a Raffles Institution alumnus with a Berkeley degree and stints at Microsoft and LinkedIn, returned to Singapore for an AI PhD at NUS and decided that ChatGPT's rise left the 'other 99.5%' of internet users — people without PCs or premium AI budgets — unserved.
Agnes launched on Product Hunt on July 4, 2025 and grew fast: 2 million registered users and roughly 150,000 daily actives within four months, about half from Southeast Asia, according to TNGlobal. By December 2025 the company reported 3 million users and 200,000 DAU, top-10 Google Play productivity rankings in the Philippines, Vietnam and Indonesia, and active fundraising at a valuation above $100 million.
On the technical side, Agnes trained its own 7B model, Agnes-R1, claimed state-of-the-art results on reasoning benchmarks (34.1% average improvement over DeepSeek's GRPO framework per the company), and routed roughly half of traffic to self-developed models to cut token costs. It also partnered with Agora for real-time AI group chat and multi-agent collaboration, supporting minority languages that global LLMs overlook.
In February 2026 Agnes closed a $10 million Series A led by LOOK FORWARD VCC, and by April 2026 the company said the wider ecosystem served nearly 10 million users and was approaching $20 million ARR, with developer API services planned for Q2 2026 and expansion into Latin America and the Middle East. Growth figures are company-reported.
What has to be true
- Yang targeted the 'other 99.5%' of internet users — Android-first, low-income, non-English speakers that OpenAI and Perplexity don't serve — instead of competing head-on for pro users.
- A fully proprietary stack trained and hosted in Singapore let Agnes sell neutrality and local-language capability in markets wary of US and Chinese AI.
- Product simplicity — search, research, slides and design in one app — lowered the barrier for first-time AI users and drove word-of-mouth growth.
- Cost efficiency (about 90% of top-tier capability at roughly 10% of the cost, per Yang) made free and cheap tiers sustainable in low-ARPU markets.
What can be applied
The wedge isn't just a cheaper model — it's choosing customers the incumbents ignore (Android-first, non-English, low ability to pay), then building product and model around them.
Aftermath
As of September 2026 Agnes AI continues to scale from Singapore: after the February 2026 $10M Series A, the company reported in April 2026 that the broader ecosystem was approaching 10 million users and $20M ARR, with developer API services planned. It is expanding into Latin America and the Middle East, training larger regional models with NUS and NTU, and previewed a 'Vibepod' AI headset with Agora. User, valuation and ARR figures are company-reported.
Sources
- Singapore's Agnes AI Reaches 3M Users and Launches Proprietary SOTA Models
- 不做通用Agent：新加坡全民AI应用Agnes，上线四月揽获20万日活用户
- Singapore's Agnes AI partners Agora to launch real-time AI workspace
- Singapore's own AI company Agnes AI Expands AI Infrastructure Platform and Confirms Series A Funding
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card