A new evaluation of the DRACO benchmark's Brave Search group shows that GPT-5.6-sol has entered the rankings with an average score of 63.99 and a cost of $1.71 per task. However, an integrated solution called OpenSquilla 0.5.0 Preview, utilizing multiple domestic Chinese models, achieved a slightly higher score of 64.09 at a significantly lower cost of $0.12 per task. This suggests a shift in complex agent task competition from relying on a single powerful model to effectively organizing multiple models. AI
IMPACT Highlights the growing importance of model orchestration over single-model supremacy for complex agent tasks.
RANK_REASON The item reports on benchmark results for AI models, specifically comparing performance and cost on the DRACO benchmark. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
