Inception’s page opens with today, and the product date at the end is Oct 19, 2026. The page prints no clock. The model announced is Mercury 3, a diffusion language model. The company says that, across a set of benchmarks, it outperforms GPT-6 Luna and Claude Haiku 5.5, and that it runs 4 to 8 times faster at more than 1,000 tokens per second. The parameter count is written as more than 40 times Mercury 1. Another sentence says intelligence on a range of standard benchmarks rose 3 times from Mercury 1 mini to Mercury 3. Speed on NVIDIA GPUs is written as 1,068 tokens per second. Context is written as 1 million tokens, against 260,000 for Mercury 2.5. The page does not attach the name of each benchmark or the other model’s score.

[1]
A flat charcoal GPU module with a muted blue edge
A GPU module lying flat, the top painted charcoal, with a muted blue edge and a row of contacts at one end., AI-generated illustration, not a news photograph

Web navigation has one result with its conditions written down. The company says that, inside its own production web-agent harness, on 50 live Mind2Web tasks, Mercury 3 completed 50%, significantly more than GPT-6.1-Sol, using the same token volume and the same number of agent steps as a frontier model. The other completion rate is not printed. “Significantly more” cannot be turned into a point gap.

[1]

The prices are dollar figures on the page, not a conversion. List price is $0.20 per million input tokens and $0.70 per million output tokens. The launch discount is written as half off: $0.10 input and $0.35 output. Cached tokens are 90% below the input price. The page says the model is available the same day on Inception’s API and through OpenRouter, Baseten, and Vercel. The 0.9-second median generation time and the 2.2-second end-to-end time in the Octen quotation are that company’s own test, not a leaderboard published by Inception.

[1]