Models

Status

What Sage Is Actually Built For

Every lineup of AI models faces the same fundamental design tension. You can optimize a model to handle a wide range of tasks reasonably well, producing something that is never the worst choice but also never the best one for anything in particular. Or you can build models with distinct, deliberately different priorities, accepting that each one will be genuinely excellent at a specific category of work and less suited to others.

Kira Sage is the second kind of model. It is being built as our most capable option in the Kira family, specifically designed for the hardest category of work: deep reasoning across complex problems, long multi-step tasks that unfold over many dependent actions, and nuanced writing where the exact phrasing carries real weight and a close-enough answer is not actually close enough.

When quality matters more than speed, Sage will be the model you reach for, not because it is the newest or the most talked about option in the lineup, but because the task genuinely deserves the extra depth and the extra depth will genuinely change the result.

It is worth being specific about why Sage exists as its own distinct model rather than simply being a slower, more thorough version of something that already exists. A model tuned to handle everything reasonably well tends to be the best choice for nothing in particular. It is a compromise built to avoid being bad at any one thing rather than being genuinely excellent at the thing that matters most in a specific moment. Deep research, careful multi-step planning, and nuanced writing all share something meaningful in common: getting them right depends on patience, thoroughness, and a genuine willingness to work through complexity rather than arriving at a fast answer that sounds plausible but cuts corners in the reasoning behind it.

Sage is being built specifically around that requirement, not as an afterthought added to a faster model, but as its own distinct approach to reasoning from the ground up.

What Sage Is Designed For

Complex Research and Analysis

This is the kind of work where a shortcut in reasoning does not just produce a slightly worse answer. It produces a wrong one, and the cost of that is almost always higher than the cost of waiting a bit longer for something genuinely dependable. When you need to understand something deeply, evaluate competing sources of information, synthesize a nuanced picture from complex material, or produce analysis that someone else will rely on to make a real decision, the quality of the reasoning behind the output matters enormously. Sage is being built for exactly this category of work, where getting the details right is not optional and thoroughness is not a nice-to-have.

Multi-Step Planning and Long Agentic Tasks

Some work does not resolve in a single exchange. It unfolds over a sequence of steps where what happens in step four depends on what actually occurred in step two, and where losing track of an earlier detail partway through does not just produce a weaker result, it can undermine the coherence of everything that follows. Coordinating a sequence of dependent actions requires holding onto context across a genuinely long piece of work and reasoning through consequences in a way that faster, lighter-weight models genuinely struggle with at the edges of their capability. Sage is being built to handle this kind of sustained, sequential reasoning reliably rather than approximately.

Careful and Nuanced Drafting and Editing

A quick draft is perfectly adequate for a routine reply. But there is a meaningful category of writing where the exact phrasing carries real consequences. A difficult message to a client where the wrong tone creates a problem that the right tone would have avoided. A carefully reasoned proposal where the precision of the argument determines whether it lands. A piece of communication where subtle distinctions in word choice change the meaning in ways that matter to the outcome. For this kind of writing, a model that produces the first reasonable-sounding version is not the right tool. Sage is being built to actually sit with the nuance rather than surface past it.

What to Expect From Sage

The Highest Reasoning Quality in the Kira Family

This is not a marginal improvement over other models in the lineup measured in small percentage points on a benchmark. It reflects a genuinely different set of priorities in how the model is built and tuned, prioritizing depth, thoroughness, and reliability in complex situations over the speed and efficiency that make other models in the family the right choice for everyday work.

Strong Adherence to Detailed and Multi-Part Instructions

A complex request with multiple specific conditions gets handled the way you actually specified it, not a simplified approximation that technically addresses the surface of the request while quietly ignoring the parts that were harder to satisfy. If you give Sage five specific conditions to account for, the expectation is that it genuinely accounts for all five rather than treating the two easiest ones as a representative sample of the whole request.

Reliability Across Long and Context-Heavy Tasks

A research task spanning many pages of source material. A plan with dependencies stretching across dozens of sequential steps. A document that needs to maintain internal consistency across a large amount of content. These are situations where a model needs to hold onto earlier details across a genuinely long piece of work without losing track of them as the task progresses. Sage is being built specifically to handle this kind of sustained attention reliably, because shorter and faster interactions simply do not test this capability in the way that real complex work does.

When Sage Will Be the Right Choice and When It Will Not

We would rather be direct about this now than let you assume that the most capable model is automatically the right choice for every task, because that assumption leads to using a slower, heavier tool for work that did not need it and waiting longer for a result that would have been equally good from a faster model.

For a quick scheduling confirmation, a routine email reply, a straightforward question with a clear answer, or any of the dozens of everyday tasks that make up most of a working day, the extra depth Sage provides will not meaningfully change the outcome. It will simply mean waiting longer for a result that another model in the Kira family could have delivered just as well, with no real difference in quality that would justify the wait.

Sage is genuinely built for the specific moments when depth and reasoning quality actually change the result. A complex research task where getting the nuance right matters and getting it wrong has real consequences. A piece of writing where the exact phrasing carries weight beyond what a quick draft can reliably produce. A multi-step plan with genuine stakes riding on the quality of the reasoning rather than just the speed of getting to an answer. These are the moments Sage is being built for, and knowing when a task is actually one of those moments is just as important as having the capability available when it is.

The practical guidance we would offer once Sage is available mirrors advice that holds true across most AI systems with multiple model tiers. Start with what the task actually seems to need. Reach for Sage specifically when you encounter a genuine requirement for depth, not as a default habit applied to everything regardless of whether the work calls for it.

Availability

Kira Sage is currently in active development. Detailed specifications including context window size, benchmark performance, and pricing will be published closer to launch when those details are finalized and verified rather than projected.

What is described on this page is directional rather than final. It is an accurate description of what Sage is being built toward and what priorities are shaping its development, not a locked-in commitment to exact specifications this far ahead of release. As with any model still in development, some details may shift as the work progresses and real-world testing surfaces things that refine the approach.

• Follow launch news→
• Request early access→

What Is Available Today

While Sage is in development, Kira is fully available and powering the full range of platform capabilities today including email management through Email Agent, calls through AI Call Agents, scheduling through Smart Calendar and Meetings, knowledge management, integrations, and everything else across the platform.

Sage will expand your options when it launches by adding a model specifically suited to the deepest and most complex work the platform supports. It will not replace what already works well for everyday tasks. It will give you a more capable option for the moments when everyday is not quite enough.