Opus training format shift: From plain text to XML tags
Ratter · x · 2026-09-01
A follow-up to the previous post. The author notes that Opus models (since around version 4.8) seem to have changed training formats, moving from plain text \n\nHuman: / \n\nAssistant: to using XML tags.
New weak evidence suggests the model sometimes perceives message boundaries as the word 'ant', as in <antml:blah>..., indicating the use of tags like <antml>.
Related event: Users Say Opus Training Shift to XML Tags Explains Leak Behavior(3 posts)→
More from Models
- OpenAI Codex sees exponential growth amid Anthropic user decline — FinanceYF5 · 2026-09-01
- Meme Suggests AI Models Are Distilling Each Other — toomanynamesaretook · 2026-09-01
- Hypothesis on Opus 5 leakiness: Mixed old and new training formats — Ratter · 2026-09-01
- Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored Trends on HF — DavidAU · 2026-09-01
- Matrix-Game 3.5: Enhancing Real-Time Streaming Interactive World Models with Patch Memory — Runjia Qian · 2026-09-01
- Qwen Team Analyzes Qwen3.8-Next Architecture Design — Qwen · 2026-09-01