Which models does the company Perplexity provide on their platform as of 02-04-2024?
- User
Which models does the company Perplexity provide on their platform as of 02-04-2024?
Assistant
Based on the information available around April 2024, Perplexity offered the following models:
For Free Users:
- The default model was based on OpenAI's GPT-3.5, fine-tuned by Perplexity for speed and web browsing [1, 10].
- Access to Perplexity's own models, such as
sonar-small-chat,sonar-small-online,sonar-medium-chat, andsonar-medium-onlinewere also available via API and potentially within the platform [9, 12].
For Perplexity Pro Subscribers ($20/month):
- GPT-4: Provided access to OpenAI's more powerful model [10, 13, 15][1].
- Claude 3: Access to Anthropic's Claude 3 model family (like Opus and Sonnet) was available [10, 13, 15][1][2].
- Experimental Perplexity Models: Access to models developed or fine-tuned by Perplexity [9].
- Open Source Models: Via the API, users could access models like Llama 3 (e.g.,
llama-3-8b-instruct,llama-3-70b-instruct) and Mistral models (e.g.,mistral-7b-instruct,mixtral-8x7b-instruct,mixtral-8x22b-instruct) [9, 12]. Note that some of these, like specific Mistral and older Llama models, were scheduled for deprecation shortly after April 2024 [12].
Key points from around April 2024:
- Perplexity Pro offered a choice between several advanced third-party models and potentially experimental Perplexity models [10, 13, 15].
- The free tier primarily used a fine-tuned version of GPT-3.5 [1, 10][3].
- Perplexity's own "Sonar" models were being introduced around this time [9, 12].
Current Models (as of early 2025 for context): Perplexity Pro now includes newer models like GPT-4o, Claude 3.5 Sonnet (and Haiku/Opus versions), Llama 3.1 based Sonar models (Small and Large), Grok-2, Gemini 2.0 Flash, and specialized reasoning models like OpenAI's O3-mini and DeepSeek's r1 [1, 2, 3, 4]. The free tier may still use a version based on GPT-3.5 or a similar capable model [1, 3][3].
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.