Inception has launched Mercury 3, the fourth iteration in its series of diffusion-based large language models. The company claims the model achieves speeds exceeding 1,000 tokens per second. Mercury 3 features a parameter count over 40 times larger than the initial Mercury 1 mini model, while benchmark quality has improved approximately 3 times.
In performance testing on medium settings, Mercury 3 scored 76.84 on IFBench and 51.08 on SWE-Bench Atlas. The model outperformed GPT-6 Luna and Haiku 5.5 in these specific benchmarks. However, it trailed competitors in tests such as DeepSWE v1 and MRCR v2, with long-context retrieval and extended coding tasks identified as current limitations.
The model is available via the Inception API Platform, OpenRouter, Baseten, and Vercel, utilizing an OpenAI-compatible API. The standard pricing is $0.20 per million input tokens and $0.70 per million output tokens, with a 50 percent discount currently applied at launch. New users receive 100 million free tokens, and enterprise options include dedicated capacity and autoscaling.
