In brief
- Mercury 3 was announced on October 19.
- Speed figures and comparisons come from the developer.
- Evaluate total time, accuracy and the need for corrections.
What Inception announced
Inception introduced Mercury 3 on October 19. Its new diffusion language model is available through the company's API and on OpenRouter, Baseten and Vercel, according to the official announcement. The company reports a context window of one million tokens and throughput of 1,068 tokens per second on NVIDIA GPUs.
The announced base rates are 0.20 dollars per million input tokens and 0.70 dollars per million output tokens, with a 50% launch discount. These are the provider's rates and may change. Its speed and quality comparisons are also Inception's own claims, not the results of independent testing by En Foco.
Faster answers do not always mean faster results
Consider a request that involves finding information, reviewing several pages and then writing an explanation. Generating the visible text is only part of the process. Measuring that speed alone does not tell you how long the whole task took or whether the result needed correcting.
To evaluate a fast tool, we suggest starting with a familiar task. Time it from the moment you send the request until you have something usable, and note any corrections you need to make. An instant answer that forces you to start over can be less useful than a slightly slower one that gets it right.
Three questions for a simple test
First, does it preserve the names, figures and conditions you gave it? Second, does it recognize when information is missing? Third, do the time savings hold up when you repeat the exercise with different inputs? This simple approach helps you look beyond an eye-catching number in a presentation.
For an initial comparison, all you need is nonconfidential material and an expected result you can check. You do not need to move an entire workflow or share sensitive documents to see whether the tool suits you. Mercury 3 is a new option; its value will be clearer in the work it gets done than in how quickly it fills the screen.
Sources and context
Sources used for this article. Follow the links to read the original material.
See the linked sources for more information. Corrections and contact →
