Citi sees frontier AI models widening lead over open-weight rivals

Investing.com - Citi said the capability gap between proprietary frontier AI models and open-weight alternatives is widening after a period of convergence, with the latest data showing frontier separation as regulatory and technical factors shift the competitive landscape.

The firm said the highest AA Intelligence Index score for proprietary models rose to 58 from 53, while the highest score for open models increased to 46 from 44, widening the closed-open gap to 12 points from 9 points. Private evaluations place leading open systems further behind in recent weeks, particularly on long-horizon cyber tasks and production reliability.

Citi said much of the earlier gap narrowing occurred when regulatory concerns restricted advanced model releases and open-weight developers benefited from distillation and access to advanced offshore compute. The firm said increasing compute availability should allow frontier labs to distill models more efficiently than open-weight competitors.

The firm noted that Opus 5.5 lifted the AA intelligence frontier 9% and costs 40% less than Opus 5 on typical workloads, according to Anthropic on September 22. Successive Grok and Gemini releases have improved intelligence while maintaining flat token pricing.

Citi said it expects enterprises to adopt mixed-model portfolios as a de-risking mechanism, while AI labs may increasingly embed routing and orchestration capabilities into the model layer directly.

You May Also Be Interested In

• First Merchants CEO to retire; president set to take over in 2027
• Nektar drug shows durability in skin disease trials at EADV 2026
• Aptera cuts capital estimate to reach production to $25 million