Aiven says cheaper AI models cut its own AI costs by 80 percent

Aiven, a Finland-based company that hosts databases for other businesses, says it cut its internal AI costs sharply by switching most work to cheaper models. Its strategy chief says its heaviest users now spend roughly 80 percent less. The company is now building a gateway, a kind of switchboard for AI models, and plans to offer it to customers as a service.

What Aiven did

Aiven faced a choice: set strict limits on how much AI its developers could use, or find a cheaper way to use it. It chose the second. Ville Lehto, its VP of strategy, told The Register the aim was to make switching models as easy as possible for staff. The company focused on its heaviest users, because a small group of people drives a large share of spending.

How the mix works

The setup splits the job between two kinds of models. Top models from companies like OpenAI and Anthropic handle the planning. Cheaper open-weight models do most of the actual tasks. Open-weight models have publicly released internals, so they can be run locally or hosted at lower cost.

• Top models plan the most critical parts of the work.
• Open-weight models do the majority of the tasks.
• Examples include Meta's Llama, Alibaba's Qwen, DeepSeek, Mistral and Google's Gemma.
• Aiven runs the models itself on rented GPUs, the chips that power AI.

Aiven built the gateway on LiteLLM, an open source framework. It added budgeting, access management and other open source parts.

Why it matters to businesses

Many firms pay for AI by usage, so bills can climb fast when a few people use it constantly. Aiven's approach suggests another route besides hard caps: let cheaper models carry the routine work and save the expensive ones for planning. Lehto also noted that models differ in quality, so cost per task is a better measure than price alone.

What is not known yet

The 80 percent figure applies to Aiven's heaviest internal users, not to every company or every worker. The article gives no price or launch date for the gateway service. It also does not say how well customers will see the same savings. Aiven is likely to be one of several vendors pitching ways to lower AI costs.