Inception Labs released Mercury 2.5 on September 8, 2026, targeting apps that need fast replies from a language model, including voice assistants, search and coding agents.
Mercury generates text through diffusion: it refines multiple pieces of a response rather than producing each token strictly one after another. Inception reports 1,107 output tokens per second on its NVIDIA GPU setup. That figure measures output generation, not the completion time of a task using multiple model calls and tools.
The API offers a 260,000-token context window, configurable reasoning, parallel tool calls and JSON output aligned with a requested schema. Inception lists standard pricing at $0.20 per million input tokens and $0.75 per million output tokens. Its launch offer reduced those rates by 80%, but the announcement does not establish how long the promotion will last.
Mercury 2.5 is available through the Inception API, Baseten and OpenRouter. The company also announced previews of Mercury Voice and Mercury Router.