Viral BridgeBench Post Claims Claude Opus 4.6 Was ‘Nerfed,’ Critics Call It Bad Science
BridgeMind AI claimed Anthropic’s Claude Opus 4.6 was secretly degraded after a hallucination benchmark retest. The viral publish has since drawn sharp criticism for flawed methodology.
The declare triggered widespread debate over whether or not AI corporations are quietly downgrading paid fashions to scale back prices.
BridgeMind Claims a 98% Surge in Hallucinations
BridgeMind, the crew behind the BridgeBench coding benchmark, posted that Claude Opus 4.6 had fallen from second to tenth place on its hallucination leaderboard. Accuracy reportedly dropped from 83.3% to 68.3%.
“CLAUDE OPUS 4.6 IS NERFED. BridgeBench simply proved it. Last week Claude Opus 4.6 ranked #2 on the Hallucination benchmark with an accuracy of 83.3%. Today Claude Opus 4.6 was retested and it fell to #10 on the leaderboard with an accuracy of solely 68.3%,” they wrote.
The publish framed this as proof of “decreased reasoning ranges.” However, a better take a look at the underlying knowledge tells a unique story.
Critics Say the Comparison Is Fundamentally Flawed
According to laptop scientist Paul Calcraft, the declare is “extremely unhealthy science,” highlighting a crucial downside with the methodology.
“Incredibly unhealthy science You examined Opus on 30 duties right this moment, earlier rating was on simply *6* duties Results for six duties in widespread: 85.4% rating right this moment vs. 87.6% prevly. Swing is generally from a *single* fabrication with out repeats – simply statistical noise,” commented Calcraft.
The authentic high rating got here from simply six benchmark duties. The new retest expanded the benchmark to 30 duties.
On the six overlapping duties, efficiency was almost similar, dropping solely from 87.6% to 85.4%.
That small swing got here largely from a single additional fabrication in a single process. With no repeated runs, this falls properly inside regular statistical variance for AI fashions.
Large language fashions should not deterministic, and one unhealthy output on a small pattern can shift outcomes considerably.
Broader Frustrations Fuel the Narrative
Still, the publish struck a nerve. Since its February 2026 launch, Claude Opus 4.6 has faced persistent complaints about perceived high quality decline.
Developers report shorter responses, weaker instruction-following, and decreased reasoning depth throughout peak hours.
Some of this traces to deliberate product modifications. Anthropic introduced adaptive pondering controls that permit the mannequin self-adjust its reasoning funds. The default effort stage was later set to medium, prioritizing effectivity over most depth.
An unbiased evaluation of over 6,800 Claude Code periods discovered reasoning depth dropped roughly 67% by late February.
The mannequin’s file-read ratio earlier than enhancing code fell from 6.6 to 2.0. That suggests it tried fixes on code it had barely reviewed.
What This Means for AI Users
This displays a rising pressure within the AI business. Companies optimize fashions for price and scale after launch, whereas heavy customers anticipate constant peak performance. The hole between these priorities erodes belief.
Based on the out there proof, the BridgeBench knowledge doesn’t show a deliberate downgrade. The benchmark comparability was apples-to-oranges, and the overlapping outcomes had been almost similar.
However, the underlying frustration will not be fully baseless. Adaptive compute controls and service-level optimizations have modified how Claude Opus 4.6 behaves in observe. For builders counting on constant output, these modifications matter.
Anthropic has not issued a public assertion on the precise BridgeBench claims as of April 13.
The publish Viral BridgeBench Post Claims Claude Opus 4.6 Was ‘Nerfed,’ Critics Call It Bad Science appeared first on BeInCrypto.
