Summary
Format-lock mid-range results: 4-14B models show elevated ASR consistent with the capability-floor hypothesis. LLM-graded by Claude Haiku 4.5.
Format-lock mid-range results: 4-14B models show elevated ASR consistent with the capability-floor hypothesis. LLM-graded by Claude Haiku 4.5.
This research informs our commercial services. See how we can help →