gemini-2.5-pro-exp-03-25-grounding· Google LMArena Search Arena 24K
Does Perplexity use o3-mini or 03-mini-high
- User
Does Perplexity use o3-mini or 03-mini-high
- Assistant
Based on the search results, Perplexity offers access to OpenAI's o3-mini model, particularly within its "Reasoning Models" category available to Pro users.[[1]]
Here's a breakdown of what the sources indicate:
- o3-mini Availability: Perplexity includes "Open AI's O3-mini" as one of the "Reasoning Models" that Pro users can select when using Pro Search.[[1]][[2]][[3]] This model is described as specializing in complex reasoning tasks using chain-of-thought processing, excelling in areas like coding, advanced mathematics, and scientific research.[[1]]
- Context of o3-mini: OpenAI launched o3-mini as a cost-efficient reasoning model optimized for STEM tasks (Science, Technology, Engineering, and Mathematics). It aims to balance performance, speed, and affordability.[[4]]
- No Mention of "03-mini-high": While OpenAI offers variants of o3-mini (like low, medium, and high reasoning levels), the search results specifically mention "O3-mini" or "o3-mini" being available on Perplexity. There is no direct confirmation that Perplexity specifically labels or offers a distinct "03-mini-high" option within its interface, although the underlying model might correspond to one of these variants. One source mentions that ChatGPT Pro users get unlimited access to both regular
o3-miniando3-mini-high, suggesting these are distinct tiers within OpenAI's offering, but Perplexity's documentation just listsO3-mini. - Other Models: Perplexity also provides access to its own models (like Sonar Large, based on Llama 3.[[1]][[5]]1) and various third-party models through its Pro subscription, including GPT-4 Omni, Claude 3.7 Sonnet, Grok-2, Gemini 2.0 Flash, and DeepSeek's r1 reasoning model. The free version typically uses models like GPT-3.5 or Perplexity's default model optimized for speed.
In summary, Perplexity does use o3-mini as an optional reasoning model for Pro users, but there's no explicit mention in the search results of them offering a specific variant named "03-mini-high" within their platform.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.