When AI bills spiral out of control, model routers become the new favorite for enterprises to reduce costs.

```

As the cost of enterprise AI usage continues to rise, a technology called “model routers” is rapidly moving from a marginal tool to the mainstream. These systems can automatically allocate the most suitable AI model based on task complexity, significantly reducing expenses without a substantial sacrifice in quality. They are attracting broad attention from startups to large enterprises.

The core logic of model routers is: not every task requires the most expensive, cutting-edge model. Basic tasks—such as summarizing emails or searching documents—can be handled by open-source models or older proprietary models, at a fraction of the cost of top-tier models. Companies like Snowflake and Palo Alto Networks have confirmed to The Information that using cheaper models for specific tasks has led to considerable cost savings.

This trend is generating real business returns. Construction company McCarthy Building reported that, using Palantir’s routing tool Evolve, their AI token usage dropped by 60% compared to the same period last year. Palantir itself disclosed that in one specific case, the tool switched tasks from OpenAI’s GPT-5.1 to the smaller GPT-5.4 Nano model, slashing compute costs by 97%.

From Manual Model Selection to Automatic Routing: An Industry Turning Point

The concept of model routers is not entirely new, but it truly entered the public eye after OpenAI released GPT-5. This model automatically switches among different models inside ChatGPT based on the complexity of user prompts, with the routing logic embedded in the product. Subsequently, routers capable of scheduling models across multiple providers have quickly spread.

Routers in the market come in various forms: independent products, built-in modules from cloud service providers, and customized solutions created by enterprise IT departments. The common goal of these tools is to replace manual model selection, thereby maintaining output quality while lowering costs.

Databricks’ Unity AI Gateway is one example. CEO Ali Ghodsi commented that this tool is “very popular,” largely because many companies are “exhausting their budgets at too rapid a pace.” Databricks had used the product internally for some time before launching it to customers.

From Startups to Tech Giants: Full-Scale Entry

The router field is attracting participants of all sizes. According to a previous report by The Information, in April this year, routing technology startup OpenRouter completed a new $120 million round of financing, reflecting capital market enthusiasm for this direction.

OpenRouter’s “automatic router” decides which model to call based on the user’s preferences for cost and quality (set on a scale from 0 to 10). Data shows that about a third of the time, the router chooses Google’s relatively inexpensive Gemini 2.5 Flash Lite, while the more powerful OpenAI GPT-5.5 is used only about 10% of the time. The automatic router is powered by startup Not Diamond, which focuses on developing routing systems for AI programming agents.

Japanese AI lab Sakana AI recently launched a multi-model collaboration system based on routers. In testing, the system mainly assigns math problems to OpenAI’s GPT-5.5 and science tasks to Google’s Gemini, because it judges these two models as better in their respective fields. Sakana AI claims the overall system performs about the same as Anthropic’s Fable 5 and Mythos Preview models in benchmarks on programming, engineering, science tasks, and reasoning.

AI programming app Cognition also released a new router this week, using its internal benchmarks to identify the relative strengths of different agents and introducing a “sidekick” agent to handle simpler tasks. Cognition says the router reaches Fable 5-level scores on a programming benchmark, but with costs 35% lower.

DIY Routing: Low-Cost Solutions Also Work

Not all enterprises need to buy professional routing products. Developers can use AI programming agents like Claude Code to build their own routers, or even let an AI model decide which model is best suited for a particular query.

Hunter Bown, responsible for AI agents at Arcee AI, said he often uses DeepSeek V4 Flash for model selection because of its low cost. His method is to provide DeepSeek with a list of models and let it judge which best handles the current prompt.

However, these “quick build” solutions also have limitations. Martian founder Shriyash Upadhyay pointed out that more complex routers sometimes perform well in benchmark tests but may not align in real-world performance. He also noted that even sophisticated routers find it difficult to accurately predict the best model just from a user’s first prompt.

Upadhyay said the rapid iteration and changing capabilities of models make routing decisions more complicated. “Companies don’t have infinite data on all different tasks, so you really have to dig deep into the models themselves to figure out what they’re good at.” For this reason, Martian not only considers the output results but also the internal computational processes of the models when making routing decisions.

Ongoing Cost Pressure, Router Demand Expected to Grow

Enterprise anxiety over AI costs is not a short-term phenomenon. As employee usage of advanced AI models (the “tokenmaxxing” phenomenon) continues to grow, management’s scrutiny of AI spending is intensifying. This backdrop provides a continuous driver of demand for model routers.

Palantir’s Evolve tool, in addition to routing functionality, can automatically adjust prompt content based on the selected model and prevent repeated requests sent to the model—a common cause of overcharges. The McCarthy Building case shows that by optimizing prompt structure, companies can consume fewer tokens when using advanced models while obtaining the same output results.

For investors, the heating up of the model router field means: on one hand, startups focused on routing technology like OpenRouter are gaining favor from capital; on the other, companies like Databricks and Palantir, which integrate routing functions into enterprise AI platforms, are strengthening their product competitiveness. As AI infrastructure spending continues to expand, the tool layer that helps companies control this cost is becoming an emerging market that cannot be ignored.

Risk Warning and Disclaimer ClauseThe market involves risks; investment needs caution. This article does not constitute personal investment advice and does not take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their specific situation. Investment based on this is at your own risk. ```