Developers building voice or real-time apps where speed matters most should test this fast diffusion model at modest token prices.
Kenning's take
What it does
Mercury 2 is a reasoning language model from Inception that generates text by diffusion, refining many tokens at once instead of one at a time. Inception says this lets it decode over 1,000 tokens per second on standard NVIDIA GPUs, so it can reason and still answer within the half-second budget of a phone conversation. Developers use it through the Inception API as mercury-2, and it is also on Azure AI Foundry; Inception released Mercury 2.5 as its successor in September 2026.
- Diffusion language model with reasoning
- 1,000+ tokens per second on NVIDIA GPUs
- Built for voice agents, search and coding subagents
- Available on Azure AI Foundry
- OpenAI-style chat completions API
Who it suits. Developers building voice agents, search pipelines and other apps where response speed matters.
Key facts
- Category
- AI models
- Made by
- Inception
- Released
- 24 February 2026
- How to get it
- through the Inception API (model mercury-2) and Azure AI Foundry
- Works on
- API
- API price
- $0.25 per million input tokens, $0.75 per million output
Models and features
- Mercury 2.5 (Inception) 2026-09-08
FAQ
When was Mercury 2 released?
Inception introduced Mercury 2 on 24 February 2026.
What makes Mercury 2 fast?
It is a diffusion model that refines many tokens in parallel, which Inception says gives over 1,000 tokens per second on standard NVIDIA GPUs.
How much does Mercury 2 cost?
$0.25 per million input tokens and $0.75 per million output tokens.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.