All Perplexity models available in 2025: complete list with Sonar family, GPT-5, Claude, Gemini and more.

Graziano Stefanelli
Aug 10
4 min read

Perplexity now includes its own Sonar models and a powerful selection of external LLMs.

As of August 2025, Perplexity has evolved into one of the most flexible AI platforms by offering a dual-layered model system. On one side, it deploys its proprietary Sonar family of models for fast, grounded, and deep responses. On the other, it allows Pro and Max users to select from a curated set of third-party models, including GPT-5, Claude Opus 4.1, Gemini 2.5 Pro, and Grok 4, unlocking advanced capabilities across search, reasoning, and research.

This structure makes Perplexity a modular tool for users looking to balance speed, accuracy, and depth of reasoning, depending on the task at hand. Below is a full breakdown of all models currently available, their differences, and when and why to use them.

The Sonar family powers Perplexity’s core response system.

The Sonar models are proprietary LLMs developed and fine-tuned by Perplexity itself. They handle the default experience for free users and serve as the foundation for hybrid tasks such as live search, web-grounded Q&A, and deep research. Each Sonar variant is optimized for a specific workflow.

● sonar

This is the default lightweight model, designed for fast answers with real-time citations. It’s ideal for everyday queries where response speed and clarity are more important than depth or complexity. It grounds almost every reply with fresh sources pulled from the web.

● sonar-pro

An enhanced version of the standard Sonar model. It executes more precise searches, delivers richer context, and handles multi-layered follow-ups more effectively. This model is triggered in more demanding sessions (e.g. Pro/Max workflows) or through long-form querying.

● sonar-reasoning

This version adds Chain-of-Thought capabilities and is better suited for logical reasoning, evaluations, or tasks that require maintaining multiple pieces of information over several steps. It operates with embedded live search and supports more nuanced decision-making.

● sonar-reasoning-pro

An upgraded reasoning model powered partly by DeepSeek-R1 1776, a custom variant of DeepSeek’s open model integrated by Perplexity. It’s especially effective in workflows that require step-by-step inference and comparison, such as research synthesis and evaluation tasks.

● sonar-deep-research

The most advanced Sonar model, used specifically for Deep Research sessions. It can autonomously conduct multi-query research threads, generate source-rich structured reports, and synthesize long documents. Often used for business intelligence, legal research, and market tracking.

Pro and Max plans unlock access to GPT-5, Claude, Gemini, Grok, and DeepSeek.

While the Sonar family covers most free and default workflows, users on Perplexity Pro ($20/month) or Perplexity Max ($40/month) gain access to a growing list of high-end external models. These include both OpenAI and Anthropic systems, along with Google, xAI, and others. Some are reserved only for Max-tier users.

● GPT-5 (OpenAI)

Available to Pro and Max users, this is the most recent version of OpenAI’s model family. It offers general-purpose excellence across all tasks, especially complex prompt chaining, creative writing, and synthesis. In Perplexity, it powers certain advanced "Pro Search" features when manually selected.

● GPT-4 Omni (GPT-4o)

Still available as an option, this model is widely used for faster general-purpose tasks that require strong accuracy but not necessarily the full reasoning power of GPT-5. GPT-4o remains popular for its multimodal capacity, although Perplexity currently limits its interaction to text-only.

● o3 and o3-pro (OpenAI)

OpenAI's lighter GPT-4 variants. The standard o3 is occasionally used for fast completions, while o3-pro is included for Max users who want cost-efficient, high-quality responses in logic-heavy workflows. These models bridge speed and coherence.

● Claude Sonnet 4.0 and Claude 4.1 Opus Thinking (Anthropic)

Anthropic’s Claude models shine in reasoning and long-document analysis. Sonnet 4.0 is available to Pro users, while Claude 4.1 Opus Thinking is exclusive to Max. The “Thinking” variants are optimized to maintain structure, summarize PDFs, and simulate cognitive tasks such as ethical evaluations, legal summaries, and structured Q&A over large contexts.

● Gemini 2.5 Pro (Google)

Included in Pro and Max plans, Gemini is strong at multi-turn conversations, code explanations, and structured logic with grounding. While it doesn’t match Claude in terms of document understanding or GPT-5 in flexibility, Gemini remains a reliable alternative with a large context window.

● Grok 4 (xAI / Elon Musk)

This model is geared toward users interested in non-mainstream analysis, Twitter/X discussions, and sarcastic but insightful reasoning. It’s not the most accurate model, but its personality-driven responses make it suitable for cultural, political, and real-world commentary.

● R1 1776 (DeepSeek variant)

Not shown as a selectable model but used as backend for the reasoning engine in sonar-reasoning-pro. This variant enables logical chaining and enhanced internal coherence, particularly for technical or mathematical problems.

● Sonar Large (Llama 3.1 70B fine-tuned)

A Perplexity-custom baseline model, available in some versions of the app for testing and fallback. It's a fast, clean model, used when external model selection is disabled or unnecessary.

Choosing the right model in Perplexity depends on your goal.

Perplexity users now have one of the widest model selections available in a single AI platform, with seamless transitions between internal and third-party engines. Here's how to decide:

For quick factual answers with citations → use sonar or sonar-pro
For logical breakdowns or planning → choose sonar-reasoning-pro or Claude Thinking
For creative tasks, coding, or multi-modal work → switch to GPT-5 or GPT-4o
For structured longform research → activate Deep Research with sonar-deep-research
For culture, current events, or tech analysis → try Grok 4 or Gemini 2.5 Pro

While the free version of Perplexity gives access to fast and high-quality models, only Pro and Max unlock the full spectrum of reasoning, file handling, and advanced workflows.

Let me know if you'd like a version of this with comparison tables, costs per API, or model latency benchmarks.

____________

DATA STUDIOS

datastudios.org