"Lands next week" (it did)

Subject: Mercury 3

Claim

"The most intelligent diffusion LLM ever released." It "outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec."

Finding

Announced as landing "next week", it landed the following week, a delivery rate we would like to see more of. The quality comparison is the lab's; the second source is a customer's "internal benchmarks". No independent index score had been published at the time of this build. One public bake-off reported about 488 tok/s and an empty answer to a coding prompt.

Timeline

15 Oct 2026
"Our largest and most capable model lands next week."@StefanoErmon · ♥ 14 ↗
19 Oct 2026
It lands.@StefanoErmon · ♥ 5.3K ↗
19 Oct 2026
"In our internal benchmarks, Mercury 3 blasted through the pareto frontier."@lluismanrique · ♥ 41 ↗

Primary sources

The posts this entry is based on, embedded as published.

Stefano Ermon

@StefanoErmon

Our largest and most capable model lands next week. If you’re building something where speed matters, we’d love to get it in your hands early: inceptionlabs.ai/enterprise#con… (2/3)

15 Oct 2026♥ 14Open on X ↗

Stefano Ermon

@StefanoErmon

1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.

19 Oct 2026♥ 5.3KOpen on X ↗

Luis Manrique

@lluismanrique

we're stoked to bring a new SOTA diffusion model to @orchestradotai customers. in our internal benchmarks, Mercury 3 blasted through the pareto frontier for efficiency

19 Oct 2026♥ 41Open on X ↗