The era of the monolithic AI model is over for enterprises. Data from Perplexity Enterprise, used by 92% of Fortune 500 companies, reveals a significant shift towards what the company calls 'enterprise AI model switching.' Instead of relying on a single go-to AI, businesses are increasingly using a diverse fleet of models, each suited to specific tasks and teams.
The Fragmentation of AI Usage
By late 2025, usage data showed a stark departure from earlier trends. While two top models dominated enterprise queries at the start of the year, by December, four distinct models each held over 10% of the market share. The leading model captured only about 23% of queries, indicating a highly fragmented landscape where no single AI reigns supreme.
This fragmentation is driven by a growing sophistication in how companies deploy AI. Beyond engineering teams, which show specialized preferences, other departments demonstrate a more varied approach. This suggests a move away from standardization and towards a more agile, task-specific AI strategy.
The Rise of the 'Multi-Model Workforce'
Employees are now treating AI models as a utility belt, selecting the right tool for the job. Perplexity data shows that 12.5% of enterprise users qualify as 'active power users,' engaging with AI at least 12 days a month. These users are at the forefront of model exploration, with 40% actively using six or more different AI models, compared to 20% of regular users.
This trend scales significantly with larger accounts. The top 50 enterprise clients utilize an average of 30 different AI models, a stark contrast to the seven models used by typical accounts. This indicates that large organizations are actively integrating a wide spectrum of AI capabilities into their operations.
Matching Models to Tasks
The rapid release of new AI models by leading companies, with Perplexity integrating 46 new models in 2025 within 24 hours, fuels this trend. This aggregator approach simplifies adoption, allowing employees to switch between models without managing multiple logins or contracts. This seamless access is critical for enabling enterprise AI model switching.
Across the board, 43.6% of organizations used more than one AI model in 2025. A notable 9.1% of users employed multiple models on a single day, highlighting the practice of routing different tasks to specialized AIs. For those who actively choose their models, 53% switched between them at least once within a single workday, underscoring how routine this 'model-hopping' has become.
Model Specialization by Use Case
While programming remains an exception where Anthropic’s Claude models see significant adoption (38% of programming queries), other functions show less clear dominance. Preferences are highly task-dependent:
- Visual Arts: Gemini Flash (40%)
- Financial Analysis: Gemini 3.0 Pro Thinking (31%)
- Debugging: Claude Sonnet 4.5 (30%)
- Software Development: Claude Sonnet 4.5 (29%)
- Legal/Court Cases: Claude Thinking models (23%)
- Medical Research: GPT-5.1 Thinking (13%)
These preferences are fluid. As new models emerge, leadership in specific task areas will inevitably shift, reinforcing the need for adaptable access.
The Evolving AI Landscape
The model ecosystem is in constant flux. At the start of 2025, Claude Sonnet 4 and GPT-4o jointly accounted for over 91% of enterprise queries. By year-end, this distribution had broadened significantly: Gemini 3.0 Pro Thinking led with 23.3%, followed by Claude Sonnet 4.5 (20.6%) and GPT-5 variants. New models experience brief spikes in usage upon release, quickly tapering off as users integrate them into their workflows or move on to newer iterations.
This dynamic environment underscores the strategic advantage of flexibility. Relying on a single 'best' model is untenable when the landscape changes weekly. Access to a comprehensive suite of models, allowing teams to pick and choose based on current needs, is the new enterprise imperative.
