Skip to content
Perplexity

Sonar Deep Research

Sonar Deep Research is a large language model made by Perplexity, added to the catalog on 7 Mar 2025. It has a context window of 128,000 tokens and costs $2.00 per 1M input tokens and $8.00 per 1M output tokens. It accepts text and produces text. Typical uses: generate comprehensive reports in finance, conduct in-depth technology research and analyze complex health topics.

Price checked · source: OpenRouter

What Sonar Deep Research is

Perplexity: Sonar Deep Research, released in March 2025, is a text-based model designed to handle complex research tasks by autonomously retrieving, synthesizing, and reasoning across diverse topics. With a substantial context length of 128,000 tokens, it efficiently processes input text to generate comprehensive reports. The model's unique ability to conduct multi-step retrieval and evaluation makes it suitable for domains such as finance, technology, and health. It supports various parameters, including frequency_penalty and temperature, to refine its output. This model excels in generating detailed insights by integrating and reasoning through extensive research material.

Key features

  • Multi-step retrieval and synthesis
  • Autonomous source evaluation
  • Extensive context length of 128,000 tokens
  • Supports diverse parameters for customization
  • Text-to-text input and output modalities

Typical uses

  • Generate comprehensive reports in finance
  • Conduct in-depth technology research
  • Analyze complex health topics
  • Stay updated with current events
  • Refine research approach autonomously

Sonar Deep Research pricing

US dollars, list price through OpenRouter.

Checked
Sonar Deep Research price list
ItemStandard
Input, per 1M tokens$2.00
Output, per 1M tokens$8.00
Web search, per search$0.005

No price change since it was added on 7 Mar 2025.

What typical tasks cost with Sonar Deep Research
What a task costsCost
Answer a short question500 in / 300 out$0.0034
Summarise a 50-page PDF25,000 in / 800 out$0.056
Review a 2,000-line source file30,000 in / 2,000 out$0.076
Read a 300-page contract set150,000 in / 1,500 out$0.312
Check the provider's pageToken counts are estimates. A maker's direct price and its monthly plans can differ from the per-token price shown here.

Sonar Deep Research specifications

Sonar Deep Research limits and identity
Model idperplexity/sonar-deep-research
Versionsonar-deep-research
Context window128,000 tokensAbout 96,000 English words
Max output115,200 tokens
Knowledge cutoffNot published
Added to catalog7 Mar 2025
Sonar Deep Research capabilities
Input typestext
Output typestext
Tool callingNo
Structured outputsNo
Reasoning effort controlNo
Content moderationNot applied by provider

Sample outputs

The same prompts are given to every model, and the unedited output is shown with the date it was produced. Upend does not publish leaderboard scores it did not run.

Write a short essay

RUN20 Nov 2026
Prompt

Write 150 words on how AI might positively upend work, leisure and creativity

Output

# How Artificial Intelligence Might Positively Upend Work, Leisure, and Creativity Artificial intelligence represents one of the most transformative forces of our time, with the potential to fundamentally reshape how we work, spend our free time, and express ourselves creatively. Rather than replacing human potential, AI is emerging as a powerful amplifier of human capability, freeing individuals from routine tasks to focus on what makes us uniquely human: strategic thinking, emotional connection, and creative innovation. This comprehensive report explores how AI can catalyze positive change across three interrelated domains—work, leisure, and creativity—examining the mechanisms through which this transformation occurs, the evidence supporting these possibilities, and the critical conditions necessary for realizing AI's most beneficial potential. ## AI's Transformative Role in the Future of Work The relationship between artificial intelligence and employment represents one of the most consequential shifts in economic history. Rather than viewing AI purely as a threat to employment, emerging evidence suggests that when deployed thoughtfully, AI can create an environment where work becomes more meaningful, productive, and aligned with human potential. The transformation occurring in workplaces today demonstrates that AI is creating a new paradigm of human-machine collaboration that elevates rather than diminishes human contribution. ### Productivity Gains and the Liberation of Human Potential At the most fundamental level, AI dramatically enhances worker productivity by automating routine tasks that consume significant portions of the workday. Research from the St. Louis Federal Reserve found that among workers using generative AI in the previous week, users saved an average of 5.4% of work hours, translating to approximately 2.2 hours per week for someone working 40 hours[7]. More specialized studies reveal even more dramatic gains: business professionals using AI could write 59% more business documents per hour, while programmers using AI assistance could code 126% more projects per week[17]. Support agents equipped with AI assistance handled 13.8% more customer inquiries per hour[17]. However, the significance of these productivity gains extends far beyond numerical efficiency metrics. When AI handles data entry, email management, routine report generation, and other repetitive administrative tasks, it fundamentally liberates human workers to engage in activities that require distinctly human capabilities[9]. This reallocation of time represents what researchers describe as a "shift left" approach to work design—moving workers away from tasks that machines perform better toward tasks where human judgment, creativity, and emotional intelligence provide irreplaceable value. Workers are thus freed to engage in strategic problem-solving, relationship building, complex analysis, and innovation that drives genuine organizational value[1]. This productivity enhancement proves particularly beneficial for less experienced or lower-skilled workers. Research from Harvard Business School and Boston Consulting Group involving over 700 consultants found that consultants scoring below average in their work experienced a 43% improvement in decision-making quality when using AI, compared to only a 17% improvement for above-average performers[61]. This democratizing effect suggests that AI can function as an equalizer, reducing the performance gap between novices and experts by handling the heavy cognitive lifting of data processing and analysis, thereby freeing working memory for creative and strategic thinking[17]. ### Job Creation and the Emergence of New Roles While concerns about technological unemployment persist, substantial evidence indicates that AI will generate employment growth alongside displacement. The World Economic Forum's Future of Jobs Report 2025 projects that while AI will displace 92 million jobs by 2030, it will simultaneously create 170 million new roles[62][65]. This net job creation reflects a fundamental economic principle: productivity improvements that reduce the cost of goods and services increase demand, which in turn drives the need for new workers to meet that expanded demand[1]. More specifically, analysis of nearly 180 million global job postings from 2023 to 2025 reveals that machine learning engineers represent the single largest growing job category, with postings surging 40% from 2024 to 2025[25]. Beyond machine learning engineering, the entire AI infrastructure stack is experiencing explosive growth: robotics engineers, research and applied scientists in tech, and data center engineers all show strong double-digit growth rates[25]. Additionally, emerging categories such as AI ethics specialists, human-AI collaboration designers, and physical AI specialists are creating entirely new professional niches[65]. These new roles require skills that build upon rather than replace traditional expertise. The jobs being created increasingly demand individuals who can combine domain-specific expertise with AI literacy—professionals who understand both the technical capabilities of AI systems and the substantive domains in which those systems operate[65]. This creates an opportunity for career evolution where experienced workers transition from performing routine tasks within their domain to serving as expert guides for AI implementation within that domain, creating higher-value roles than what existed previously. ### Reducing Cognitive Burden and Enhancing Well-being One of the most underappreciated benefits of AI in the workplace relates to cognitive burden and burnout prevention. Cognitive load theory, developed by educational psychologist John Sweller, identifies that human working memory has finite capacity, and when cognitive load exceeds that capacity, performance deteriorates and stress increases[51]. In modern workplaces, particularly for knowledge workers, information overload and complex task management frequently exceed cognitive capacity, contributing substantially to burnout[51]. AI addresses this challenge by functioning as what researchers describe as "forklifts for the mind"—tools that handle heavy cognitive lifting so humans can focus their mental resources on higher-order thinking[17]. By automating data aggregation, information synthesis, and routine decision-making, AI reduces extraneous cognitive load, allowing workers to direct their limited cognitive resources toward genuinely complex problems, strategic thinking, and creative innovation[55]. For neurodivergent individuals managing attention difficulties, executive function challenges, or sensory processing variations, this cognitive support proves particularly transformative[51]. The well-being implications extend further. By automating drudgery, AI provides workers with greater autonomy over their time and attention. Research grounded in self-determination theory—which identifies autonomy, competence, and relatedness as fundamental human needs for thriving—suggests that AI can satisfy all three needs[74]. Automating tedious tasks gives workers autonomy over how they spend their time; immediate feedback and suggestions build competence; and by freeing space for deeper collaboration, AI strengthens the relatedness essential for meaningful work[74]. Studies show that knowledge workers receiving substantial AI training saved at least 11 hours weekly and reported significantly higher engagement and satisfaction[74]. ### The Caveat: Extended Work Hours and Value Capture However, emerging research reveals a significant counterpoint to the optimistic productivity narrative. A study by economists at Emory, Auburn, Fordham, and Seton Hall universities examining data from the American Time Use Survey found that workers in occupations with higher exposure to generative AI experienced a significant increase in work hours and a decrease in leisure time following the introduction of ChatGPT[3]. Based on data from 2022-2023, an interquartile increase in generative AI exposure corresponded to an additional 3.15 hours of work and a reduction of 3.20 hours in leisure per week[3]. This phenomenon reflects what the researchers term the "extended workday" problem: while AI increases productivity, the benefits of that increased productivity accrue primarily to employers and consumers rather than workers themselves[3]. In competitive labor markets where workers have limited bargaining power, productivity gains from AI do not necessarily translate into reduced working hours or increased compensation; instead, companies extract that surplus value by requiring workers to accomplish more within the same or longer timeframes[3]. Furthermore, AI surveillance technologies that monitor remote workers enforce greater productivity, with the result being longer hours coupled with lower job satisfaction despite higher wages[3]. This critical finding suggests that realizing AI's positive potential for work requires deliberate policy and organizational choices. Without conscious intervention to ensure that productivity gains benefit workers alongside employers, AI risks becoming a tool for intensification rather than liberation—enabling companies to extract more value from workers rather than freeing workers from drudgery. Organizations committed to positive AI implementation must consciously design systems that translate productivity gains into improved work-life balance, enhanced compensation, or meaningful career development rather than simply demanding more output. ## The Creative Renaissance: AI Enhancing Artistic Expression and Execution While some observers fear that AI will render human creativity obsolete, evidence from both research and practice demonstrates that generative AI functions most powerfully as a creative partner that enhances rather than replaces human artistic capability. This partnership appears particularly transformative in democratizing creative access and accelerating the execution phase of creative work. ### Enhancing Individual Creativity While Democratizing Access Research from a randomized controlled study published in Science Advances examined how access to generative AI affects human creativity in storytelling[6][54]. The study found that access to generative AI ideas caused stories to be evaluated as significantly more creative, better written, and more enjoyable, particularly among less creative writers[6][54]. Writers in the "human with one generative AI idea" condition showed an increase in novelty of 4.7% compared to writers without AI access, while those with access to five AI ideas showed an 8.1% increase in novelty[6][54]. Similarly, AI access increased the usefulness of stories by 9% for those with access to five ideas compared to those without AI[6][54]. Critically, the research found that generative AI particularly benefited writers with lower baseline creative ability[6][54]. This democratizing effect reflects a broader pattern observed across professional domains: AI tends to reduce performance gaps between high-skill and low-skill individuals by providing capabilities that less experienced people lack. Access to AI doesn't make everyone equally creative, but it does enable individuals without innate creative genius to produce work of surprising quality and originality. The mechanism underlying this enhancement reflects how generative AI functions as an idea amplifier. Rather than replacing the creative process, AI generates numerous variations and possibilities that humans then evaluate, refine, and synthesize into a final creative work[2][8]. The iterative loop between human judgment and AI generation—where AI suggests possibilities and humans assess which suggestions advance their creative vision—proves particularly powerful because it combines AI's ability to generate vast quantities of options with human aesthetic judgment and intentionality[27]. Across creative disciplines, this pattern repeats. Musicians find that AI-generated melodic and harmonic suggestions accelerate composition while preserving artistic intention[2][8]. Visual artists report that AI-generated variations and patterns serve as starting points for more complex designs that benefit from human refinement[2][8]. Writers describe AI as eliminating friction in knowledge retrieval, enabling them to produce finished works in record time[8]. Designers note that AI handles tedious formatting tasks—adjusting font sizes, aligning graphics—that consume time without contributing to creative problem-solving[8][12]. ### Accelerating Creative Execution and Freeing Human Cognitive Resources One of the most underappreciated benefits of AI for creative work involves what might be called "execution acceleration"—speeding the translation of creative vision into finished work. Every creative endeavor involves necessary but tedious tasks that often function as bottlenecks constraining creative output[8][12]. Writers must research extensively, transcribe interviews, and manage citations. Designers must adjust spacing, test layouts, and format assets for multiple platforms. Musicians must handle recording logistics, mixing, and publishing requirements. AI transforms these constraints into non-issues. Writers can query knowledge bases with zero friction, retrieving information that would previously require library research[8]. Designers no longer spend hours adjusting fonts and aligning elements—AI assists with these technical aspects, freeing designers to focus on conceptual innovation[8]. Photographers can instantly retouch and upscale images rather than spending hours on manual editing[8]. Musicians can generate rough instrumental accompaniments, freeing their attention for vocal performance and interpretation[8]. The practical result proves transformative. Creators with large bodies of existing work report that AI enables repurposing content at unprecedented scale[8][12]. A content creator who previously hired professionals to convert their YouTube videos into blog posts reports that AI-generated conversions exceed the quality of human-written versions, completing in seconds what previously required paid professional labor[8]. This capability doesn't replace the creator's creative judgment—someone still must select which content to repurpose and evaluate whether AI conversions require refinement—but it dramatically expands what individual creators can accomplish. This execution acceleration proves particularly valuable because it liberates the cognitive and emotional resources that manual tasks consume. Researchers distinguish between the cognitive burden of execution tasks and the creative energy required for conceptual work[8]. By automating execution, AI redirects human attention toward the higher-order creative work that requires artistic vision, emotional depth, and intentional meaning-making[8][12][24][28]. ### Power Democratization and New Economic Possibilities Perhaps the most economically significant benefit of AI for creative work involves fundamentally shifting power dynamics. Historically, the tools required for professional-quality creative production—recording equipment, design software, publishing platforms—represented substantial capital investments accessible primarily to well-funded organizations and wealthy individuals. This created barriers to entry that limited creative opportunity to those with financial resources or institutional access. AI-powered tools democratize these capabilities. An individual with no design background can now generate professional-quality graphics using AI image generators. A solo musician can create full orchestral arrangements without orchestrating live musicians or hiring expensive arrangers. A writer without publishing connections can generate polished marketing materials that rival outputs from professional marketing firms. This democratization doesn't mean everyone becomes equally skilled, but it does mean that technical expertise becomes less limiting than intentionality and artistic vision[12][24][28]. The economic implications prove significant. Individuals and small businesses can now accomplish work that previously required hiring expensive specialized professionals or maintaining in-house teams[12]. This reduces barriers to entry for entrepreneurs and creative professionals operating independently, enabling more people to participate in creative economy opportunities[12]. While some traditional intermediaries (editing services, design firms, production companies) face pressure from this democratization, the expansion of creative opportunity and the reduction of barriers to creative participation represents a net positive for creative expression overall. However, this democratization requires one critical condition: AI must function as a complement to human creativity rather than a replacement for it. Individuals who leverage AI to complement their abilities—using it to amplify their creative vision while applying their judgment and aesthetic sensibility—develop what might be called "creative superpowers," accomplishing work they could not achieve independently[12]. Conversely, individuals who attempt to use AI as a substitute for creative thinking, accepting AI outputs without applying meaningful judgment, produce derivative work that lacks authenticity and originality[23]. ### The Collective Creativity Question and Cautionary Notes Despite the substantial evidence for AI-enhanced individual creativity, researchers note an important potential concern regarding collective creativity. While individual writers using AI generate stories evaluated as more creative, there exists a troubling dynamic: if writers discover that AI-influenced work receives higher creative evaluations, they have incentive to use AI more frequently, potentially reducing collective novelty as more creators converge toward AI-suggested ideas[6][54]. This represents a genuine tension between individual and collective creative benefit that societies may need to actively manage. Additionally, concerns persist regarding over-reliance on AI stifling skill development. Artists who rely excessively on AI tools without continuing to develop foundational skills risk atrophying their abilities[16]. The most concerning scenario involves creators mistaking AI convenience for creative mastery, accepting AI outputs without applying artistic judgment, and subsequently producing work that lacks authenticity and emotional depth[23]. These risks suggest that optimal AI integration in creative work requires conscious cultivation of what might be termed "creative discernment"—the ability to evaluate AI-generated ideas critically and apply human judgment to identify which suggestions genuinely advance artistic vision. ## Reimagining Leisure: Personalization and Co-Creation in Entertainment As AI transforms how we work, it simultaneously offers profound possibilities for reimagining leisure—the time away from paid work that humans dedicate to recreation, rest, personal development, and pleasure. The transformation of leisure from passive consumption to personalized, interactive, co-created experiences represents one of AI's most overlooked positive potentials. ### From Passive Consumption to Personalized Co-Creation Traditional leisure has often been characterized by passive consumption: individuals watch films selected by studios, listen to music curated by record companies, read books determined by publishing industries. While these experiences offer genuine value, they reflect a one-directional model where consumers receive predetermined content with minimal personalization. AI fundamentally restructures this model by enabling hyper-personalized leisure experiences adapted in real-time to individual preferences[4][20][26]. Recommendation systems learn individual taste profiles through behavioral observation—what g

Sonar Deep Research questions

How much does Sonar Deep Research cost?

Sonar Deep Research costs $2.00 per 1M input tokens and $8.00 per 1M output tokens through OpenRouter, as checked on 22 Nov 2026. Summarising a 50-page PDF costs about $0.056.

What is the context window of Sonar Deep Research?

128,000 tokens, or roughly 96,000 English words. It can write up to 115,200 tokens in one response.

Can Sonar Deep Research read images and PDFs?

No. It accepts text only. It produces text, so it cannot generate images.

How do I try Sonar Deep Research?

Use the "Try Sonar Deep Research" button on this page. It opens the chat with Sonar Deep Research selected; you type the question and press send.

Try Sonar Deep Research on your own question.

It opens in the chat with this model already selected.

Sign in to Upend

Use an account you already have, or get a one-time code by email.

Use a password instead

For accounts that already have a password. Forgotten it? An emailed code signs you in just the same.