May 2025 Summaries

Perplexity has launched Labs for Pro subscribers, a new mode designed to turn ideas into completed project deliverables such as reports, spreadsheets, dashboards, marketing materials, and simple web applications. Unlike Perplexity Search, which focuses on quick answers, and Research, formerly Deep Research, which provides in-depth responses in roughly three to four minutes, Labs can spend 10 minutes or more conducting self-supervised work using deep web browsing, code execution, and chart and image generation. Labs can write and run code to organize data, apply formulas, and create files, while collecting generated assets for viewing and downloading within the workflow. It can also build and deploy basic interactive mini apps directly in the interface. The feature is available through the mode selector on the web, iOS, and Android, with Mac and Windows support planned, and Perplexity encourages users to explore examples in its Projects Gallery or apply Labs to a personal or professional task.

Perplexity has announced a global partnership with seven-time Formula One World Champion Lewis Hamilton, linking the company’s focus on fast, accurate knowledge discovery with Hamilton’s reputation for precision, discipline, and continual learning. Hamilton describes curiosity as essential to success in sport and life, saying that he uses Perplexity because the best performers never stop asking questions. The collaboration emphasizes shared values of speed, understanding, and purposeful progress, positioning Perplexity as a platform trusted by millions to turn questions into informed action while reflecting Hamilton’s approach to decision-making both on and off the track.

Perplexity is introducing two updates intended to expand subscription value and improve access to trusted information: Pro Perks, which offers Pro subscribers deals from brands in travel, health, finance, and other categories, and premium content integrations with Statista, PitchBook, and Wiley. The company says the new data partnerships will enable answers to cite professional statistical, market, financial, and healthcare sources that are often limited to enterprise customers, with access allocated through Premium searches: three monthly searches for free users, five for Pro subscribers, and ten for Enterprise Pro clients. Perplexity presents both initiatives as part of its effort to make AI-supported decision-making more useful, accessible, accurate, and trustworthy, while planning to add further offers and data sources over time.

Perplexity and SAP announced at SAP Sapphire 2025 that Perplexity’s answer engine will be integrated into SAP Joule, enabling users to receive AI-generated answers directly within SAP workflows. The collaboration combines Perplexity’s contextual search capabilities with SAP’s Knowledge Graph and secure enterprise data to provide structured metrics, charts, graphs, and scenario-based insights grounded in both real-time business information and relevant external events. The companies emphasize accuracy, privacy, data ownership, and enterprise standards, positioning the integration as a way to reduce complexity in organizational data and help workers make more informed business decisions at scale.

Perplexity and Wiley have partnered to integrate Wiley’s scientific, technical, medical, and educational materials into Perplexity’s answer engine for institutions using Enterprise Pro subscriptions. The integration allows students to access assigned textbooks and other curriculum materials without switching platforms, ask questions about course content, generate study guides, receive tailored explanations, and connect academic concepts with current web-based examples. Educators can use the tools to create customized learning resources, prepare lessons more efficiently, and incorporate real-time developments into curricula. The partnership also aims to promote responsible AI literacy through factual accuracy, trusted citations, attribution, and critical evaluation of information. Institutions can enable the connection by obtaining Wiley authorization and linking their Wiley account through Perplexity Enterprise Pro administration.

The report describes efforts to develop reinforcement-learning infrastructure and train mathematical reasoning models using Group Relative Policy Optimization (GRPO), finding that RL can substantially improve performance beyond supervised fine-tuning when models begin with strong long chain-of-thought capabilities. The team selected NVIDIA’s NeMo Aligner as a short-term platform because of its feature completeness and support, while planning a longer-term move to the simpler, internally maintainable torchtune framework; it also improved rollout throughput with vLLM integration and resolved major log-probability alignment, memory, and implementation issues. Experiments on Llama 3.1 8B models and math datasets including GSM8K, MATH, ORZ, NuminaMath, and MATH-500 showed that an ORZ-supervised warm-up followed by GRPO achieved roughly 0.70 pass@1 on MATH-500, compared with about 0.38 for the original instruct model and 0.60 for the SFT model. Results indicated that varied training difficulty, appropriate learning rates, moderate sampling temperature, and prior exposure to long reasoning traces strongly affect training efficiency and stability. The work also identified self-repetition and maximum-length collapse as important failure modes, especially when format rewards or overly aggressive learning rates were used, while noting that GRPO’s group-based advantage normalization can create zero-gradient cases and favor easier questions. Future work will examine curriculum-learning strategies, revised reward and advantage calculations, and continued infrastructure and algorithm improvements.