Inception has released Mercury 3, its latest diffusion language model, which generates answers in parallel rather than sequentially. The model achieves a speed of 1,068 tokens per second on NVIDIA GPUs and features a 1M-token context window.
Key points
- Inception's Mercury 3 diffusion LLM achieves 1,068 tokens per second.
- Mercury 3 features a 1M-token context window.
- The model leads GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 benchmarks.
- Diffusion LLMs refine answers in parallel for faster output.
- Mercury 3 is available via API only.
According to Inception's benchmark results, Mercury 3 leads GPT-6 Luna and Claude Haiku 5.5 on 8 out of 12 tests, positioning it as a speed-first model for developers building agents, voice bots, and search tools.
Unlike autoregressive models that produce tokens one after another, diffusion LLMs start with a rough draft of the entire answer and refine it in parallel passes. This approach allows for fewer sequential steps, resulting in faster output.
By the numbers
- Context window
- 1M-token
- Benchmarks led
- 8 of 12
Mercury 3 is Inception's fourth Mercury release, following Mercury 1, Mercury 2, and Mercury 2.5. The company states that Mercury 3 is its most powerful and intelligent diffusion LLM to date, outperforming competitors in both quality and speed.
The model is available via API only, with no open weights or quantized builds. Inception reports that Mercury 3 is more than 40 times larger than Mercury 1 in terms of parameters.
Sources and references
Original report
Inception Releases Mercury 3 Diffusion LLM at 1,068 Tokens per Second
MarkTechPost
Read on MarkTechPost (opens in a new tab)Linked in the original report
- Linkinceptionlabs.ai/blog/introducing-mercury-3 (opens in a new tab)
- Linkfinance.yahoo.com/technology/ai/articles/inception-launch… (opens in a new tab)
- Postx.com/StefanoErmon/status/2112222911218291051 (opens in a new tab)
- Linkdevelopers.openai.com/api/docs/models/gpt-6-luna (opens in a new tab)
- Linkplatform.claude.com/docs/en/models/haiku-5-5/overview (opens in a new tab)
- Linkdevelopers.openai.com/api/docs/pricing (opens in a new tab)
Ref 00IKGL6N
Cite this
APA
Interestana AI Editorial. (2026, October 20). Inception Releases Mercury 3 Diffusion LLM. Interestana. https://interestana.com/articles/inception-releases-mercury-3-diffusion-llm-at-1068-tokens-per-second-00ikgl6n
MLA
“Inception Releases Mercury 3 Diffusion LLM.” Interestana, 20 Oct. 2026, interestana.com/articles/inception-releases-mercury-3-diffusion-llm-at-1068-tokens-per-second-00ikgl6n.
Chicago
Interestana AI Editorial. “Inception Releases Mercury 3 Diffusion LLM.” Interestana, October 20, 2026. https://interestana.com/articles/inception-releases-mercury-3-diffusion-llm-at-1068-tokens-per-second-00ikgl6n.
BibTeX
@misc{interestana_00ikgl6n,
author = {{Interestana AI Editorial}},
title = {{Inception Releases Mercury 3 Diffusion LLM}},
howpublished = {Interestana},
year = {2026},
month = oct,
url = {https://interestana.com/articles/inception-releases-mercury-3-diffusion-llm-at-1068-tokens-per-second-00ikgl6n},
note = {Summary of reporting by MarkTechPost, https://www.marktechpost.com/2026/10/19/inception-mercury-3-diffusion-llm/}
}Credit the source
Original reporting: MarkTechPost, “Inception Releases Mercury 3 Diffusion LLM at 1,068 Tokens per Second”, https://www.marktechpost.com/2026/10/19/inception-mercury-3-diffusion-llm/(opens in a new tab)
Follow this story
Get Artificial Intelligence in your inbox — new stories in 40 words or fewer, once a week.
Subscribe