11X cheaper than ChatGPT: Tiny 150M model just proved AI doesn't need to "think out loud" to be smart
Pathway’s 150M model achieves 29.5% on ARC-AGI-1 while costing 11 times less than ChatGPT’s comparable reasoning model during inference.
AI lab Pathway has unveiled a new 150 million parameter model called BDH-CQ that proves advanced AI reasoning does not require expensive intermediate text generation. Released this week, BDH-CQ scored 29.5% pass@2 on the ARC-AGI-1 benchmark, which is about eleven times cheaper than the comparable ChatGPT model in terms of compute cost.
While ChatGPT's GPT 5.6 Luna model cost around $0.008 per task, BDH-CQ's inference cost was a mere $0.0007. This efficiency gain stems from BDH-CQ's unique architecture, which solves problems internally in memory instead of generating intermediate text. This internal reasoning approach could enable more cost-effective AI reasoning for real-world applications.
Amazon Web Services sees BDH-CQ as a promising step toward affordable advanced reasoning. However, current top performers like Claude Opus 5 and Gemini 3.1 Pro carry a higher price tag of around $0.5 - $0.6 per task. Pathway plans to extend BDH-CQ's approach to more challenging benchmarks like mathematical reasoning, ARC-AGI-2, and ARC-AGI-3 evaluations.
Written by urgent.news from TechRadar's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
