Introducing Mercury Voice

Voice agents need a response within about 500 ms, which forced a choice between slow reasoning models and fast non-reasoning ones. Mercury Voice reports reasoning with a 320 ms median first-token time, at $0.40/$1.50 per million tokens.

• Inception says Mercury Voice, a diffusion LLM tuned for voice agents, returns its first answer token in a median 320 ms (p95 750 ms) on production customer-service prompts, which it calls 5.9x faster than GPT-6 Luna without reasoning.
• Inception optimizes for time to first answer token, measured after reasoning finishes, because callers notice pauses beyond roughly 500 ms; it claims Mercury Voice at low effort is the only tested model with a median under that budget.
• On a composite of τ³-bench Telecom, Retail and Airline, IFBench and BFCL v4, Inception reports Mercury Voice beating GPT-6 Luna, GLM-5.3-Flash and Gemini 3.5 Flash-Lite, arguing voice builders no longer must choose fast non-reasoning models.
• The model offers low, medium and high reasoning effort, 128K context with up to 50K output tokens, and costs $0.40/$1.50 per million input/output tokens, discounted 50% at launch.

• LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
• token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
• context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
• benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

• postMercury 2 targets voice agents with a sub-300ms reasoning passInception
• postInception launches Mercury Voice, a diffusion LLM for low-latency voice agentsInception
• articleMercury 2 The First Reasoning Model Fast Enough To Pick Up The PhoneInception

Checking sign-in…

Loading comments…