Inception AI Inc. has launched its new flagship diffusion language model, Mercury 3. The company reports that the model outperformed OpenAI’s GPT-6 Luna and Anthropic’s Claude Haiku 5.5 in eight out of 12 internal benchmarks, while operating at four to eight times the speed of its competitors.
The model achieves a generation speed of 1,068 tokens per second on Nvidia GPUs, compared to approximately 240 tokens per second for Haiku 5.5 and 130 for GPT-6 Luna. Mercury 3 features a 1 million token context window, a significant increase from the 260,000 tokens supported by its predecessor, Mercury 2.5.
Pricing for Mercury 3 is set at $0.20 per million input tokens and $0.70 per million output tokens, with a 90% discount for cached input. A temporary launch promotion offers these rates at half price. Developers can access the model via Inception’s platform or through OpenRouter, Baseten, and Vercel.
Inception was founded in 2024 by Stefano Ermon, Aditya Grover, and Volodymyr Kuleshov. The startup previously raised $50 million in a round led by Menlo Ventures, with participation from Microsoft’s M12 arm. Reports from May indicated that Microsoft and SpaceX held acquisition talks with the company, valuing it at over $1 billion.
