A new model, with provider claims to assess
Inception’s October 19 announcement introduces Mercury 3, a diffusion language model. Hosted access is documented, but we have not tested it. The exact announcement hour is unconfirmed.
The provider advertises a one-million-token context window and faster output with improved benchmark results. Those performance claims are not our measurements. A large context window describes capacity, not how reliably every detail will be understood.
Current documentation lists launch API prices of $0.10 per million input tokens and $0.35 per million output tokens. These are promotional rates, not a guarantee of future pricing or total application cost.
Older-model users should check their access
Inception has closed Mercury 2 and 2.5 to newcomers. Its notice gives customers who used them before October 19 a November 15, 2026 access deadline; enterprise access follows the applicable agreement.
The source does not establish an exact cutoff hour or timezone. Do not infer automatic migration or identical replacement behavior from API compatibility.
Our take
R2’s editorial analysis: speed matters when an app repeatedly waits for a model, but a fast wrong answer still needs correcting. Compare complete tasks, tool responses, formatting, and failure handling using representative prompts.
Inventory the model names your app calls, confirm the access deadline that applies to you, and keep a recoverable configuration before changing production. Ordinary readers using unrelated chat apps need no action from this notice.