Model WarsJuly 7, 2026via MarkTechPost
Liquid AI Open-Sources Antidoom: A Final Token Preference Optimization (FTPO) Method that Reduces Doom Loops in Reasoning Models
Why it matters
Liquid AI releases a targeted solution to a critical failure mode in reasoning models—doom loops that waste context and degrade output quality. This is a meaningful technical advancement in model reliability that competitors will need to implement or match.
Key signals
- Liquid AI open-sourced Antidoom method targeting doom loops in reasoning models
- Final Token Preference Optimization (FTPO) retrains only the token position causing repetition
- LFM2.5-2.6B: doom-loop rates reduced from 10.2% to 1.4%
- Qwen3.5-4B: doom-loop rates reduced from 22.9% to 1%
- Generation, detection, and FTPO trainer components released as open source
- Addresses failure mode where models repeat token spans until context window exhaustion
The hook
Doom loops plaguing reasoning models just got solved. Liquid AI's open-source method cuts failure rates from 22.9% to 1%.
Liquid AI released Antidoom, an open-source method that targets doom loops in reasoning models. A doom loop repeats a span until the context window is exhausted. Antidoom finds the token that starts the loop and retrains only that position using Final Token Preference Optimization (FTPO). On LFM2.5-2.6B, doom-loop rates fell from 10.2% to 1.4%; on Qwen3.5-4B, from 22.9% to 1%. Generation, detection, and the FTPO trainer are open source.
The post Liquid AI Open-Sources Antidoom: A Final Token Preference Optimization (FTPO) Method that Reduces Doom Loops in Reasoning Models appeared first on MarkTechPost.