Perplexity hit a $20 billion valuation in September 2025. Revenue grew from $63 million at the end of 2024 to an estimated $500 million annualized by April 2026. I see this growth as a direct result of the product’s ability to pull users away from Google daily. The company reached $200 million in annualized revenue by October 2025. 45 million active users existed by mid-2025, and that number grew to 100 million monthly active users by April 2026.
The company raised over $1.5 billion in total funding. Key investors include NVIDIA, Jeff Bezos, SoftBank Vision Fund 2, Accel, and Databricks. In January 2024, the Series B round brought the valuation to $520 million. By December 2024, the Series D round pushed that number to $9 billion. This rapid escalation follows a trajectory where the company reached $500 million in annualized revenue in April 2026.
I recommend the Perplexity API for organizations that require high-fidelity, web-grounded answers without the overhead of managing their own search crawlers. The company manages a lean team of approximately 250 employees to generate this revenue. This capital efficiency distinguishes the company from other AI startups that burn billions on training models.
The Two-Layer Pricing Model
Perplexity API pricing involves two distinct layers: token costs and per-request fees. The first layer uses the standard LLM logic where you pay for input and output tokens. The second layer applies a fee based on how much web content the model retrieves. This request fee varies depending on the search depth and context size.
You already know that LLM providers charge for tokens, but Perplexity adds a layer of complexity that shifts the math for production-scale applications. Most LLM APIs charge per token and stop there. Perplexity charges for the intelligence and the retrieval. When developers route 50,000 queries per day through Sonar Pro at high context instead of Sonar at low context, the cost rises from $300 per day to $1,500 per day, creating a $36,000 monthly difference.
The cost per query depends on the model selection. A developer using the base Sonar model with 600 input tokens and 400 output tokens at low context will pay a different rate than a user selecting Sonar Pro at high context. The following table breaks down the base model costs for the Sonar line:
| Model | Input Tokens ($/1M) | Output Tokens ($/1M) | Request Fee ($/1K) |
|---|---|---|---|
| Sonar | $1 | $1 | $5 to $12 |
| Sonar Pro | $3 | $15 | $6 to $14 |
| Sonar Reasoning Pro | $2 | $8 | Varies |
| Sonar Deep Research | $2 | $8 | $5 (search) + $2 (cit) + $3 (reasoning) |
Deep Research Complexity
Deep Research operates as a different category of API usage. It introduces additional cost layers that do not exist in standard Sonar models. These layers include citation tokens at $2 per million and reasoning tokens at $3 per million. The model also charges $5 per 1,000 autonomous search queries.
I find the Deep Research pricing the most difficult to predict for a budget. A single query triggers multiple autonomous searches. One typical query might cost $0.30 to $1.30 depending on the depth of reasoning. The model decides how many searches to run. You do not control this number directly.
A single Deep Research query includes these components:
- Input tokens at $2 per 1M.
- Output tokens at $8 per 1M.
- Citation tokens at $2 per 1M.
- Reasoning tokens at $3 per 1M.
- Autonomous search queries at $5 per 1,000.
The reasoning tokens often drive the highest part of the bill. You pay for the model to think, search, and synthesize. This makes the Deep Research API more expensive than the standard Sonar Pro tier.
Consumer vs. Enterprise Tiers
Perplexity offers different tiers for different needs. The Free tier provides unlimited basic searches but limits Pro Search to 5 queries per day. Pro costs $20 per month and includes unlimited Pro Search and 20 Deep Research queries per day. Max costs $200 per month and provides unlimited Labs and 10,000 monthly Computer credits.
Enterprise plans cater to organizations. Enterprise Pro costs $40 per seat per month. It includes 500 Deep Research queries per day and allows for 15,000 file uploads. Enterprise Max costs $325 per seat per month. This tier provides unlimited Deep Research and unlimited Labs queries.
| Feature | Pro ($20/mo) | Enterprise Pro ($40/seat/mo) | Enterprise Max ($325/seat/mo) |
|---|---|---|---|
| Deep Research | 20 queries/day | 500 queries/day | Unlimited |
| File Uploads | 50 per Space | 15,000 org-wide | 15,000 org-wide |
| Collaboration | 5 users | 9,999 teammates | Unlimited |
| Security | Standard | SOC 2 Type II, SSO, SCIM | SOC 2 Type II, SSO, SCIM |
The Enterprise Pro plan allows for 500 Deep Research queries per day per seat. Teams can upload up to 15,000 files across all Spaces. This plan provides access to all features of the Pro tier with added administrative controls. Users benefit from Single Sign-On and SCIM provisioning for organization-wide management. The company does not train models on customer data when using this tier.
Search API and Agentic Research
Developers can choose between the Search API and the Agentic Research API. The Search API returns raw web search results without LLM rewriting. It costs $5 per 1,000 requests for standard search and $1 per 1,000 requests for the fast tier. This is a tool for builders who want to manage their own synthesis.
The Agentic Research API provides access to third-party models including OpenAI, Anthropic, Google, and xAI. These models are billed at direct provider rates. Perplexity adds a $0.005 fee per web search invocation and a $0.0005 fee for URL fetches. This provides a marketplace of models within the Perplexity infrastructure.
The Sonar API is the primary choice for most developers. It is a pay-as-you-go model with prepaid credits. The company discontinued the $5 monthly API credit for Pro subscribers. This means heavy API users must purchase credits separately.
Will Perplexity’s reliance on third-party models through Microsoft Azure become a liability if those providers decide to compete more aggressively in the search space?
Comparing Workflow Automation
Companies use different tools for different tasks. Glean serves large enterprises that want personalized workplace search across apps like Slack and Jira. It integrates with 100+ applications but lacks the workflow automation found in other tools.
Thunai focuses on enterprise AI automation through the Thunai Brain. It connects documents, emails, and CRMs to power AI voice, chat, and email agents. It is a different approach than Perplexity, which focuses on research and cited answers.
Perplexity provides a specialized research workflow. It uses a multi-model routing layer to orchestrate frontier LLMs. This allows users to compare outputs from different models in the Model Council feature. I see this as a research tool rather than a general automation engine.
| Tool | Primary Use | Best For |
|---|---|---|
| Perplexity | Cited research | Researchers and developers |
| Glean | Workplace search | Large enterprises |
| Thunai | Workflow automation | Organizations using AI agents |
Competitive Pressures and Legal Risks
Perplexity faces significant challenges from Google. Google AI Overviews directly addresses the core value of Perplexity by providing synthesized answers at the top of search results. Google has over 90% market share and massive resources to iterate on AI summaries.
Legal risks also exist. The New York Times, Dow Jones, and Reddit have filed lawsuits or sent legal notices regarding content scraping. These cases involve allegations of copyright infringement. If courts force Perplexity to change how it accesses content, the core product changes.
The infrastructure costs are a major burden. CEO Aravind Srinivas has stated that GPU procurement is a top use of capital. With 780 million monthly queries, the inference costs are massive. The company must scale revenue faster than compute costs to reach profitability.
The company is also dependent on model providers. It uses a $750 million Microsoft Azure partnership to access models from OpenAI and Anthropic. If these providers restrict access or increase pricing, the cost structure changes.
Managing API Costs in Production
Effective cost management requires monitoring token usage and request types. Developers should implement caching at the application layer to reduce the number of API calls for repeated context. This can reduce costs by 20% to 30%.
The selection of a model determines the efficiency of the budget. Routing a simple question through Sonar Pro is expensive. Using the base Sonar model for lightweight retrieval is much cheaper.
I suggest developers use the Search API for high-volume, low-complexity tasks. Use the Agentic Research API only when the task requires reasoning from models like Claude or GPT-5.
The following table shows the cost of a standard request with 600 input tokens and 400 output tokens at different context levels:
| Context Level | Total Estimated Cost |
|---|---|
| Low Context (Sonar) | $0.0059 |
| Low Context (Sonar Pro) | $0.007 |
| High Context (Sonar) | $0.012 |
| High Context (Sonar Pro) | $0.014 |
