Perplexity AI free models: releases and capabilities in 2025

In 2025, Perplexity AI significantly expanded its free and open-access model offerings, introducing new reasoning capabilities, enhanced search performance, and open-source contributions. With the launch of Sonar, the release of R1-1776, and strategic partnerships to deliver localized AI across Europe, Perplexity has positioned itself as one of the few players combining cutting-edge research with accessible, high-performance models. The September 2025 update reviews the latest releases, key features, and rollout status of Perplexity’s free AI models.

Sonar becomes the default engine for free users.

On 11 February 2025, Perplexity introduced Sonar, its in-house AI model built on Meta Llama 3.3 70B and optimized for web-grounded, real-time answers. Sonar now powers all free-tier queries, replacing legacy engines and establishing a single high-performance foundation across the platform.

Key capabilities of Sonar include:

• Context window: Up to 128,000 tokens, enabling deep document analysis.

• Web-grounded search: Integrates live data into answers for improved factual accuracy.

• Dynamic depth search: Offers High, Medium, and Low modes to balance speed and reasoning depth.

• Latency optimizations: Leveraging Cerebras wafer-scale inference, Sonar achieves sub-second response times.

By making Sonar the default model for free users, Perplexity ensures broader access to advanced multimodal capabilities without requiring a paid subscription.

Deep Research brings advanced reasoning to the free tier.

In April 2025, Perplexity integrated Deep Research into Sonar’s architecture, allowing users to perform extended multi-step analysis tasks even on the free plan, albeit with query limits. Deep Research enables the model to plan, evaluate, and synthesize results from multiple sources before responding.

Key improvements introduced:

• Better grounded reasoning across complex topics.

• Expanded citation coverage, improving transparency.

• Available via the Sonar API for developers, with pay-as-you-go scaling.

Independent evaluations place Sonar’s Deep Research mode alongside Google Gemini 2.5 Pro Grounding at the top of web-augmented benchmark scores, reinforcing Perplexity’s competitive positioning in information retrieval and synthesis.

R1-1776 marks a rare open-source release.

On 18 February 2025, Perplexity released R1-1776, a post-trained model based on DeepSeek-R1. Distributed under an Apache 2.0 licence, it became one of the most permissively open-sourced reasoning models of the year.

Model	Release date	Access	Context window	Specialization
R1-1776	18 Feb 2025	Hugging Face weights + Sonar API	128K tokens	Factual accuracy, open benchmarking
Sonar	11 Feb 2025	Default for free users	128K tokens	Web-grounded reasoning, speed optimizations

While R1-1776 was removed from hosted API availability on 1 August 2025, it remains fully accessible as an open-source model, and developers can still deploy it locally or integrate it manually. Free-tier users are now routed through Sonar Pro Reasoning, which inherits R1-1776’s analytical refinements.

Sovereign AI project introduces localized free models for Europe.

In July 2025, Perplexity announced a strategic collaboration with NVIDIA DGX Cloud to deliver Sovereign AI models, aimed at supporting 24 official EU languages while complying with AI Act data residency and transparency requirements.

Project highlights:

• Models built on custom-tuned Sonar variants for localized information retrieval.

• Hosted regionally in EU-based GPU clusters to ensure privacy compliance.

• Initial pilots launched in July, with full Sonar API integration scheduled for Q4 2025.

This initiative positions Perplexity as a pioneer in providing free, regulated AI tailored for multilingual European audiences.

Performance optimizations and reliability guarantees.

Alongside model launches, Perplexity introduced technical refinements to improve both response speed and user data privacy:

• Speculative decoding enhancements published in June reduced average latency across free queries by 30%.

• No-customer-data policy reaffirmed in updated July documentation, ensuring Sonar outputs are not used for training.

• Adaptive routing between Sonar search depths provides cost-optimized efficiency while maintaining accuracy.

These performance-driven updates aim to provide free-tier users with the same quality of reasoning and factual grounding expected from premium systems.

Overview of Perplexity’s free and open-access model releases in 2025.

Release	Date	Access method	Core features
Sonar	11 Feb 2025	Default for free users + Sonar API	Web-grounded search, 128K context, optimized latency
Deep Research integration	Apr 2025	Free queries + API	Multi-step reasoning with dynamic planning
R1-1776 (open-source)	18 Feb 2025	Hugging Face + manual deploy	Post-trained DeepSeek-R1, 128K context, Apache-2.0
Sonar performance refinements	Jul 2025	Auto-selected	Faster decoding, better energy efficiency
Sovereign AI pilots (EU)	Jul 2025	Free, regionally hosted	Localized models, 24 EU languages, AI Act-compliant

Perplexity’s positioning in September 2025.

By September 2025, Perplexity AI has secured its role as a leading provider of free, high-performance AI access. With Sonar powering the platform, R1-1776 open-sourced for research, and the upcoming rollout of localized Sovereign AI models, Perplexity balances accessibility, transparency, and scalability.

This approach differentiates Perplexity from competitors like OpenAI and Google, offering a rare combination of free advanced capabilities and open innovation, while continuing to prioritize factual accuracy and multilingual availability across markets.

____________

FOLLOW US FOR MORE.

DATA STUDIOS

datastudios.org