Proprietary frontier AI models are pulling further ahead of open-weight competitors, according to analysis from Citi released on October 2, 2026.
The performance gap on the AA Intelligence Index expanded to 12 points, with the top proprietary score rising from 53 to 58 while the best open-weight model moved from 44 to 46.
Citi said private evaluations show open-weight systems lagging further behind in recent weeks, especially on production reliability and complex, long-horizon cybersecurity tasks.
The firm pointed to Anthropic's Opus 5.5 release on September 22, which raised the intelligence frontier by 9% while cutting costs by 40% versus its predecessor.
Cost-stable updates from Grok and Gemini also featured in the shift.
Spending Falls as Adoption Climbs
Enterprise transaction data from more than 70,000 U. S.
companies tracked by the Ramp AI Index shows corporate AI spending has dropped continuously since peaking in July.
Ramp economist Ara Kharazian attributed the decline to price reductions and the rollout of cheaper, more efficient standard and lite models amid competition between OpenAI and Anthropic.
Even so, business adoption reached an all-time high by late September, with token usage surging roughly 50 percent since July.
Commercial adoption remains concentrated in proprietary models, as open-source alternatives account for less than five percent of total enterprise AI spend.
Among major corporate users, API spending data shows Anthropic leading in late September with 51 percent of token expenditure, followed by OpenAI at 44.5 percent.
To manage costs and reduce vendor dependency, Citi expects enterprises to increasingly deploy multi-model portfolios paired with integrated orchestration layers.
