- The Aurora-2 model family: Flash, base, and Reason
- 256K-token context and multimodal input
- The agent runtime and function calling
- The research, training pipeline, and safety stack
In 2026, Aurora became Qai. The models, the research, and the benchmarks all carried over. Qai is where Aurora lives now, and where you use it. If you landed here looking for Aurora, you are one click away.
Aurora started as a research project with a stubborn goal: build a model that actually reasons, and run it at a price real teams can afford.
Over three years it grew from a demo into a frontier model serving millions of people across 190 countries. It learned to see, to hear, and to hold a quarter-million tokens of context without losing the plot. It closed real bugs, passed hard exams, and traded blows with the best frontier models on the public benchmarks.
Then the work found a bigger home. In 2026, Aurora was acquired into Qai. Qai takes the Aurora-2 models, the multimodal stack, and the agent runtime, and wraps them in a product built for the people who were already leaning on Aurora every day. Same brain, sharper edges, one front door.
Head to q-ai.ca. That is the home for everything Aurora became.
Create an account for the app, or generate an API key for the models.
Aurora-2 Flash, base, and Reason are all there, now served as Qai.
Aurora AI was a frontier large language model built in Canada, first released in 2023. It handled text, images, and audio, reasoned across 256K tokens of context, and served millions of users before being acquired into Qai in 2026.
Yes, through Qai. Aurora was acquired into Qai in 2026. The Aurora-2 models still run, now delivered under the Qai name at q-ai.ca.
You use Aurora at q-ai.ca. Qai serves the full Aurora-2 model family through a single product and API.
The Aurora API moved to Qai. Point your requests at Qai and you reach the same Aurora-2 models, with the same context window and modalities.
Aurora-2 led the frontier on reasoning. It scored 89.4% on MMLU and 60.2% on GPQA Diamond, ahead of GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, and Llama 3.1 405B, and stayed within a point of the best models on code and math.
Aurora-2 scored 89.4% on MMLU, 90.1% on HumanEval, and 95.6% on GSM8K, and served 2.1T tokens per day at peak across more than 190 countries.
Everything on this site points to one place. Open Qai and start building with the models Aurora became.