Introducing Multi-Model: Keep Quality High While Cutting Token Spend

Introducing Multi-Model: Keep Quality High While Cutting Token Spend

Token spend is the new headcount problem. Your CFO is asking the same question they used to ask you about hiring: are we getting enough ROI out of what you're spending?

As a marketing leader, you know where the spend is coming from. It’s research-heavy plays, like competitive intel that pulls pricing and positioning from a dozen sources, voice-of-customer synthesis from sales calls and support tickets. It’s also production-heavy plays: first drafts and revision passes on long-form content, localized variants for every audience and locale. Not to mention the evergreen work, like sentiment tracking across LLM platforms, and campaign reporting that synthesizes performance data every week.

Each of those initiatives has a cost, and as a result, whole use cases stay off the roadmap entirely. Impactful marketing initiatives never get built because running them at the model tier they need doesn’t fit the budget.

More of this work, done well, means better outcomes for the business, but you are up against your CFO’s instinct to cut back. But what if your budget could produce more, simply by not running every task through the most expensive model available?

That's what Multi-Model makes possible. It gives you control over the model running each Playbook, so you can allocate more intelligence where you need it and spend less where you don't.

Not every marketing job needs the most expensive model

Every campaign you run has a different center of gravity and different needs. Some lean heavily on extensive internal research: pulling first-party insights from your internal systems over long periods of time. That work needs a model that is cheap and reliable.

Others require deeper reasoning: for example playing adversarial roles in brief creation process where you want different perspectives to push your thinking. That may require frontier intelligence.

Get the model wrong for the job and you're either overpaying for simple work or asking a lower-cost model to carry a job it can't handle.

AirOps helps you solve this challenge. Multi-Model gives you the flexibility to choose the specific model and reasoning effort that is the best fit for the Playbook you’re running. Don’t know what you want? Hand it to Auto Mode and let us figure it out.

How Multi-Model works

1. Choose your Playbook. In the navigation bar on the left, go to Playbooks and choose the Playbook you want to run.
2. Choose a model and reasoning effort, or keep Auto Mode. Select the model that best fits the work, then choose how much reasoning it should use. Higher reasoning effort is better suited to complex work, while lower reasoning effort keeps simpler tasks faster and less expensive. Or keep Auto Mode, which uses AirOps’ recommended default configuration, currently Opus 4.8.
3. Publish and run your Playbook. Publish your changes and run the Playbook as usual. You can return at any time and switch models in one click.
4. Tweak sub-agent models (coming soon). You can tweak the sub-agent models in your Playbook.

Quality comes from the harness, not just the model

Point a new model directly at your content challenges and you'll notice the gaps fast: brand voice that drifts by the third paragraph, citations that don't hold up, and a brief that gets lost halfway through a long piece. The model didn't get worse. It just lost the structure that was making it look good.

That structure comes from the rest of the AirOps harness, not the model itself. You can think of the harness as the environment the model lives in (much like Claude feels very different in Lovable vs Cursor vs Notion). It’s the harness that brings quality and consistency to every step of a complex marketing workflow. Built for marketing, the AirOps harness combines the instructions, tools, quality sub-agents, and checks needed to understand your brand and audience, conduct deep research, and verify the final output.

Take research and claim verification. We have tested and iterated on dedicated parts of the harness to find authoritative sources, ground factual claims in evidence, and verify those claims against the source material.

You can also measure the result for yourself. The AirOps Content Quality Score then helps you assess whether the new configuration still meets your quality bar across information gain, E-E-A-T, SERP alignment and citation likelihood. This makes it easier to balance cost, brand, quality and performance using a real signal from your own content.

Deputy saw that in practice. After moving to the new AirOps Harness, Rob Press, Content Engineer at Deputy, said: “We cut costs 64%, with output quality holding steady.”

What this looks like in practice

Your market and competitor monitoring Playbook runs leaner. Scanning sources for new launches, pricing changes, messaging updates, and industry news, then compiling the findings, can be handled by a lower-cost model with lower reasoning effort.

Your article-writing Playbook keeps its high-reasoning model. Turning a complex brief into a publishable article means balancing brand voice, audience, messaging, and structure, while verifying every claim. That work requires deeper reasoning.

If you don't want to manage any of this, turn on Auto Mode. AirOps applies its current best-fit default, and the Content Quality Score confirms nothing slipped after the switch.

Available now

Multi-Model is live in AirOps today, and currently rolling out to all customers. Support for additional models, including open-weight models, is coming soon.

Book a demo with our team today: airops.com/demo

‍