news-tldr.com

The important facts, the open questions, and the sources — without the churn.

← Latest briefing

Technology

Diffusion language model Mercury 2.5 launches with reported speed gains

Developers have released Mercury 2.5, claiming major speed and capability upgrades, according to Hacker News.

Updated Sep 8, 2026 · 21:47 UTC1 reports

The short version

  • Developers announced the release of Mercury 2.5, calling it the largest diffusion language model trained so far.[Hacker News]
  • The creators report a 40% capability gain over Mercury 2, generation speeds reaching 1,107 tokens per second, and a 260K token context window.[Hacker News]
  • Previews were also revealed for Mercury Voice and Mercury Router alongside temporary launch discounts.[Hacker News]

Key facts

  • The creators released Mercury 2.5, describing it as their most capable production model and the largest diffusion language model trained to date.[Hacker News]
  • The developers claim the model achieves 1,107 tokens per second on NVIDIA hardware with a 260K token context window and a 40% intelligence gain over Mercury 2.[Hacker News]
  • Alongside the release, the team previewed Mercury Voice and Mercury Router.[Hacker News]

What remains uncertain

  • Performance claims, latency figures, and intelligence benchmarks originate from the creators and partners without independent verification.[Hacker News]

Sources

Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.