ERNIE 5.0 Thinking Preview: benchmark results

Baidu's extended-thinking preview variant of the multimodal ERNIE 5.0 flagship. Provider: Baidu. Released 2026-01-01. Access: API.

Unified ELO 1585 ± 1, rank #305 of 1392 rated models, from 19 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA TAU-2 Bench83.92Accuracy (%)74.8
AA CritPt1.43Accuracy (%)68.1
AA Terminal-Bench Hard25Accuracy (%)66
AA Humanity's Last Exam13.25Accuracy (%)63.1
AA GPQA Diamond77.68Accuracy (%)62.9
AA Omniscience - Science, Engineering & Mathematics31.24Accuracy (%)61.9
Artificial Analysis Intelligence Index15.66Intelligence Index59.7
AA Omniscience - Law14.43Accuracy (%)58.7
AA Omniscience - Health22.2Accuracy (%)56.9
AA Omniscience - Humanities & Social Sciences22.1Accuracy (%)56.8
AA-Omniscience Accuracy22.13Accuracy (%)54.8
AA Omniscience - Business17.39Accuracy (%)52

Interactive version: theaggregate.ai/model?slug=ernie-5-0-thinking-preview · How It Works · Data refreshed daily, snapshot 2026-09-05.