ERNIE 5.0 Thinking Preview is Baidu's reasoning-focused flagship, and its benchmark profile is strong across technical dimensions: livecodebench at 0.812 and AIME-25 at 0.85 place it among the better coding and maths models in the field, while its agentic index of 54.5 and tau2 of 0.839 suggest meaningful multi-step task capability.

For businesses, the model's strengths align with coding assistance, mathematical reasoning, and agentic workflows. GPQA at 0.777 and MMLU-Pro at 0.830 reflect broad knowledge depth. However, long-context reasoning (lcr 0.067) is surprisingly weak for a model of this calibre, and no pricing data is available, making cost planning difficult.

The absence of pricing and the provider's limited presence in standard enterprise API marketplaces are practical barriers for most business teams. Organisations that can access it directly and have technical workloads to match its strengths will find it capable, but mainstream adoption will depend on broader availability.

Assessed June 30, 2026

Rankings consider pricing, capabilities, benchmarks, and real-world applicability and are refreshed as new models launch. Feedback?