Inception's Mercury diffusion LLM wows in speed test with essay on parallel thinking
gkavek · reddit · 2026-10-20
A Redditor tested the speed of Inception AI's Mercury, a diffusion-based LLM, by asking it to write a long text off its own knowledge. The model produced an essay, "On Thinking in Parallel: A Letter to the Sequential World," explaining from a first-person view how diffusion models differ from autoregressive ones — refining all tokens in parallel rather than one at a time, like sculpting with fog. On-topic, fluent, and generated rapidly, the piece doubles as a live demo of diffusion LLMs' speed advantage.
More from Models
- RealityBench Benchmarks VLMs on Documenting Physical Work, Finds Weak Spatial Reasoning — relantic · 2026-10-20
- ISTA-DASLab's NVFP4 prefiller recovers Qwen3.8-27B low-quant accuracy with SSD streaming — magnus-m · 2026-10-20
- Cursor Pro+ $60/month quota shown: Cursor Models 25% used, Other Models 21% — pramodk73 · 2026-10-20
- "OpenAI really doesn't have the capacity": model hits capacity error mid-task — cedric_chee · 2026-10-20
- DeepSeek V4.1 Flash uses 4.3x tokens and 2.4x time vs GPT-6 Sol, gap within noise — teortaxesTex · 2026-10-20
- Dev reports 5 hours of Codex outages while OpenAI status page stays all green — timigod · 2026-10-20