Every frontier held; no cluster moved and nothing was rerouted this week.

One cluster was checked this week (canaries: 1), and it held (clustersHeld: 1) with zero moved (clustersMoved: 0). The cluster is agentic tool use, a reasoning family that asks a model to chain tool calls across several steps; its stored quality was 0.97 and its observed mean was 0.95. Four items were graded (itemsGraded: 4), which is the full canary sample size; with only four points, the result is a narrow sample and its margin is wide, so a small difference like 0.02 is not a drift signal.

This week's frontiers

One row per kind of work. Routed pick is the model Potion currently sends that work to. Stored quality is its exam score when it was measured in full, give or take the margin. Canary is this week's small re-check: a few fresh tasks, scored the same way, to catch a model that has got worse. Held means the re-check landed inside the margin.

This week had one frontier cluster and it held, so no full re-measurement was started. A held verdict here means the canary mean did not fall below the stored interval's lower bound; it does not mean the observed mean sat inside the interval, and with four items the margin is wide. Next week the same cluster will be checked again with a fresh four-item canary, and the headline to watch is whether that mean stays above the stored lower bound.

Auditions

An audition is a newly released model's first exam, on the kind of work it looks suited to. It earns a place only by beating the model already doing that work on quality or price.

No new models were auditioned this week; the auditions list is empty. No candidates were screened (candidatesScreened: 0) and none were measured (candidatesMeasured: 0).

Mixing

Mixing means using two or three cheaper models together in a particular way instead of one expensive one. We score combinations by replaying results we already have, so this costs nothing to explore. How a combination works is part of the product and is not published; what it achieves is.

No model combinations were re-run this week; the mixing list is empty. Because there were no new item-level measurements to replay, no combination figures were produced and no cost band changed.

What it means for you

If you pay for AI by the request this week, the routes did not change, so the model you are routed to is the same stored pick as last week. The one check that ran was narrow — four items on one cluster — so the absence of a drift signal is a calm result, not a broad endorsement. Nothing was rerouted, and nothing new was added to the measured pool, so there is no new cost or new pick to plan around.

How the numbers are made

How these numbers are made, in plain words. We sort requests into kinds of work: sorting text into categories, pulling fields out of documents, writing code, answering from a set of documents, and so on. For each kind we keep a private exam of tasks the models have never seen. Every model sits the same exam. Code is marked by running it; other answers are marked against a reference answer. A model's quality is its average mark, and because an exam is a sample we also give a margin of error: two models whose margins overlap are called a tie. Cost is what a thousand requests would cost at the provider's public prices. A frontier is the short list of models that are the best deal at their level of quality, meaning nothing else is both better and cheaper. Each week we re-check every model on that list with a few fresh tasks to catch any that have got worse, and we give newly released models the exam for the work they look suited to.

• Every quality figure is a mean over a retrieval-hostile suite with a bootstrap 95% interval; two points whose intervals overlap are reported as tied.
• Code-generation and code-review quality is scored by executing the code; other clusters are scored by a rubric against a reference answer.
• A canary is a small weekly sample (four items) against the stored measurement; it detects collapse, not one-point movement.
• Combinations of models are replayed from stored item-level results; agreement between models is modelled conservatively, so the combination figures understate rather than overstate.

Numbers

Questions

Did any AI route change this week?

No. One cluster was checked and it held, with clustersMoved at 0, so no stored pick was replaced and no cluster was rerouted.

What does it mean when a frontier cluster holds?

In this system a held cluster means the canary mean did not fall below the stored quality interval's lower bound; a drift flag only fires on collapse, and a held cluster may still sit below its stored quality within the margin.

How many items were graded this week?

Four items were graded, all in the single canary for the agentic tool use cluster; canaries are small weekly samples that detect collapse rather than one-point movement.

Glossary. A frontier is the short list of models that are the best deal at their level of quality: nothing else is both better and cheaper. A floor is the lowest exam score you are willing to accept. A margin (or interval) is how far the true score could sit from the measured one, because an exam is a sample. A canary is a small weekly re-check. See the docs and the evidence.