Boost.ai's Presis Model Cuts Token Costs 33% in Agent Platform Push

Boost.ai's new self-hosted classification model, Presis, cut token costs by an average of 33% in internal testing and ran at least 2.5 times faster than the models it replaces. The company announced the launch on November 10, 2026, calling the speed gap most noticeable during peak-demand hours and in voice AI deployments.

Presis is not a chatbot or a generative model. It makes the routing and safety decisions inside boost.ai's agent platform: which part of an AI agent should respond, whether a question is in scope, whether an answer is safe to show an end user, and how completed conversations get reviewed. In voice deployments, those checks must finish before a response can begin, which puts classification directly on the response-time path.

The company says Presis already powers those platform decisions, with a broader rollout across the platform continuing over the coming weeks.

What do the numbers show?

Internal testing reported accuracy gains across all five classification features, ranging from 0.5 to 7.9 percentage points depending on the feature. The 33% token-cost reduction and the 2.5x speed improvement come from the same testing program.

CTO Rasmus Hauch framed the decision around infrastructure control. Owning "every part of the stack it runs on," he said, allows the classification-specific model to keep improving. Version 14 documentation places the fine-tuned models on internally managed, dedicated server clusters, with low response times and high capacity as the stated goals.

The compliance footprint stays put. Presis runs on AWS in the same region that already serves each customer instance, so the documented processing location does not change. AWS is already an approved sub-processor. That means the model introduces no new sub-processors and requires no data processing agreement change or additional agreement.

Boost.ai says end-to-end control also removes routing, scope and safeguard decisions from exposure to third-party outages, timeouts and capacity swings.

How does Version 14 split the work?

Version 14 divides the platform's language-model work into two lanes. GPT-5.1 is mandatory for built-in text-generation features and the default for new agentic actions. Presis handles the classifier component of five systems:

• Agent orchestrator
• Few shot
• AI review
• The out-of-scope detector
• System guardrails

Customers must configure GPT-5.1 before upgrading. Without it, the system migration will fail, the company says.

The architecture narrows model-provider choice for built-in classification. External Bring Your Own Model support and provider configuration options are being removed for those tasks. BYO remains available for generative Agentic Actions, LLM Actions and Guardrails.

Why the deadline is real

The transition ties directly to Microsoft's schedule. Azure retirement of GPT-4o, GPT-4.1, GPT-4.1-mini and GPT-4.1-nano lands in April 2027, and boost.ai's mandatory Version 14 changes were scheduled to start reaching production in mid-October 2026. Early migration and testing capabilities began with Version 12.8.3 back in August 2026.

Presis is the renamed Boost Native Model. Boost.ai says the Norwegian name conveys accuracy and timeliness, and that the change did not alter how the model works.

What's next?

The company is evaluating a similar dedicated-model approach for generative features. For customers, the immediate trade-off is tighter platform control and reported efficiency gains, set against less provider freedom for built-in classification. Any broader move into generation remains an evaluation rather than a committed launch — a signal that boost.ai sees self-hosted models as more than a one-off swap.