# Inception launches Mercury 3 diffusion LLM with 1,068 tokens per second

> Inception's new Mercury 3 model claims a processing speed of 1,068 tokens per second and a 1 million‑token context, topping several leading models on most benchmark tests.

Oossa · 2026-10-20 · https://oossa.com/en/inception-launches-mercury-3-diffusion-llm-with-1-068-tokens-per-second

Inception announced its latest language model, Mercury 3, on Tuesday. The company says the model can generate text at a rate of 1,068 tokens per second while handling a context window of one million tokens. In internal tests the model topped GPT‑6 Luna and Claude Haiku 5.5 on eight out of twelve benchmark suites listed by Inception.

A diffusion LLM like Mercury 3 uses a generation technique that treats text creation as a gradual refinement process, similar to how image‑generation diffusion models work. This approach is meant to improve quality and flexibility, though the company has not released independent verification of the speed claim.

## How Mercury 3 compares to other models

Inception’s own comparison table places Mercury 3 ahead of GPT‑6 Luna and Claude Haiku 5.5 on most tests. The benchmarks cover a mix of language understanding, reasoning, and generation tasks, but the article does not detail which specific tests were used or the exact scores. No third‑party evaluation has been published yet, so the results remain unconfirmed outside the company.

## The facts

- Mercury 3 processes 1,068 tokens per second, according to Inception.
- The model supports a 1 million‑token context window.
- In Inception’s table, Mercury 3 leads GPT‑6 Luna and Claude Haiku 5.5 on eight of twelve benchmarks.
- The announcement was published on 2026‑10‑20.

## Why it matters

If Mercury 3 lives up to its claims, developers could build applications that handle much longer texts without chopping them into smaller pieces, such as full‑document analysis or long‑form content creation. Faster token processing also means lower latency for real‑time services, which could make chatbots feel more responsive. However, without independent testing, it remains unclear how the model performs in everyday use compared to existing options.

## Sources & references

1. [Inception Releases Mercury 3 Diffusion LLM at 1,068 Tokens per Second](https://www.marktechpost.com/2026/10/19/inception-mercury-3-diffusion-llm/) – MarkTechPost, 2026-10-20

Last updated: 2026-10-20