Mercury Voice

Mercury Voice, released September 29, 2026 by Inception, is a proprietary voice model with 128K context and an explicit reasoning mode (low/medium/high effort) — an unusual combination, since voice models rarely expose deliberative reasoning controls. It is generally available to enterprise customers through an OpenAI-compatible endpoint, supporting up to 50K output tokens, tool use, and structured output.

The honest picture: this is an enterprise-only, closed model — no weights, no public pricing, no independent benchmarks. Inception's background is diffusion-based language models, so a reasoning voice model is a new lineage with unproven production track record, and the 'reasoning mode' adds latency to a modality where latency is the whole game.