Mercury 3 hits 2 trillion tokens in under two weeks as daily usage climbs linearly
StanfordAILab · x · 2026-11-06
Inception Labs researcher samarakhanna shared that Mercury 3, their diffusion-based LLM, has processed over 2 trillion tokens in less than two weeks since launch, with daily usage still growing linearly.
He recalls the internal excitement during training: the team used Mercury 3 daily for development work, even running multiple Mercury instances writing code and kernels to improve the model itself. He argues that scaling diffusion LLMs has clearly paid off and invites users to share what they're building.
More from Models
- ChatGPT's $20 price came from a 2023 Discord survey, and everyone copied it — aakashgupta · 2026-11-06
- ThinkyMachines' Inkling 2 Small trains models on its own: $426 of compute doubles Qwen score in 4.5 hours — Scobleizer · 2026-11-06
- Claude Max user finds limits match old Ultra tier, suspects 5x/20x tiers are gone — LeviTurk · 2026-11-06
- Gemini is bad at coding but works well as a chat mentor, says dev — lxfater · 2026-11-06
- OpenAI never launches big models on Fridays, and no Astra 6.2 teases yet — haider1 · 2026-11-06
- API dies mid-task: only Atomic Agent with local fallback aced the test, Codex just hung — Ok-Brush8377 · 2026-11-06