GPT-5.6 and Claude feel slow — where are the diffusion coding models?
viksit · x · 2026-09-03
The author complains that frontier coding models are painfully slow to generate: GPT-5.6 Sol and Claude Fable both feel sluggish, and asks where the diffusion-based coding models are — a paradigm that could sidestep the speed limits of autoregressive token-by-token generation.
More from coding & agent
- Steve Yegge: AI-built systems will eventually outgrow their own understanding—debt compounds superlinearly — teropa · 2026-09-03
- Building a Telegram Hermes agent that tracks macros, workouts and sleep — brandon_galang · 2026-09-03
- Claude Fable 5.1 one-shot builds a playable model train yard in the browser — natesiggard · 2026-09-03
- "My client thinks the agent does the work. It's me at 11pm" — automation seller's confession — 0CTAVERSE · 2026-09-03
- mitsuhiko: OpenAI's additional_tools is additive to system prompt tools, not prior message tools — mitsuhiko · 2026-09-03
- Linear's bug autofix loop closed 300+ bugs in 30 days via Datadog/Sentry and its coding agent — zeeg · 2026-09-03