The San Francisco AI company’s new Batch API pushes it further into selling infrastructure to developers, as revenue climbs and a reported funding round would value it above $30 billion.
Fifty thousand. That is how many requests a developer can now put in a single file and hand to Perplexity’s agents to work through on their own schedule, under a Batch API the San Francisco company released on Oct. 22.
The product is plumbing, not a consumer feature, and that is the point. Under Aravind Srinivas, its co-founder and chief executive, Perplexity has spent the year turning an answer engine known for citing its sources into a company that also sells the machinery underneath: an Agent API, an agent product called Computer, and now a way to run that machinery in bulk.
Mr. Srinivas earned his Ph.D. in computer science at the University of California, Berkeley, in 2021. His dissertation, “Representation Learning for Perception and Control,” filed with the electrical engineering and computer sciences department in August of that year, was written in Pieter Abbeel’s lab and dealt in part with self-attention architectures, the transformer-style designs that now underpin large language models.
What the Batch API does
According to Perplexity’s documentation, a customer uploads a JSONL file, one request per line, of up to 50,000 requests and 200 megabytes. Each request runs the full Agent API loop, so the built-in web search and page-fetching tools work inside a batch. The customer polls the job and downloads a results file when it is done.
The documentation pitches the feature for work “that does not need an immediate answer, such as evaluations, classification, enrichment, or offline research over a large dataset.” Batches aim to finish within 24 hours, according to a summary on Chasing Next, and requests that never start are not billed.
Two business details stand out. Requests inside a batch do not count against a customer’s request rate limit, though token limits still apply, which removes one of the main headaches of running large research jobs through an API. And there is no separate batch discount: “There is no separate batch price,” the documentation says. OpenAI models run on that company’s cheaper “flex” tier where available, and other models are billed at their standard rates.
The Batch and Files endpoints, the documentation notes, “follow the shape of OpenAI’s Batch API,” a choice that lowers the cost for developers of moving existing workloads. The feature is not available to customers on zero-data-retention plans, and batch inputs and outputs are kept for 30 days.
The launch follows a run of developer-facing releases. On Oct. 15, Perplexity told Semafor it was open-sourcing SPACE, the sandbox it uses to contain AI agents, with CoreWeave, Akamai and Crusoe agreeing to offer it to their customers. On Oct. 21, Mr. Srinivas said a codebase-visualization skill called Graphify was available in Perplexity Computer, with 5,000 credits offered to each user to try it, RuntimeWire reported.
The numbers behind the push
The developer products sit on top of a business that has grown quickly. Perplexity’s last closed financing, in September 2025, valued it at $20 billion, TechCrunch reported, citing The Information. The company had raised about $1.5 billion in total by then, according to PitchBook data cited in the same report.
In August, The Information reported that Nvidia, already an investor, was discussing joining a new equity round that would value Perplexity at more than $30 billion, and that the company’s annualized revenue had passed $750 million, up from less than $250 million at the start of 2026, according to TechStartups. That report attributed part of the growth to Perplexity Computer. Perplexity declined to comment on the report at the time, and no closing of the round has been announced.
At $30 billion, the reported valuation would be about 40 times that annualized revenue. Usage-billed developer products like the Batch API give the company a revenue line that does not depend on consumer subscriptions.
Not every deal has held. Snap said in May that its planned $400 million partnership to put Perplexity’s answer engine inside Snapchat had “amicably ended” in the first quarter, TechCrunch reported, and that its guidance assumed no contribution from Perplexity.
The legal backdrop
Perplexity’s agents have also been tested in court. In August, a Ninth Circuit panel vacated a preliminary injunction that Amazon had won against Perplexity’s Comet browser assistant, concluding that Amazon was unlikely to show Perplexity had “accessed” its computers under the federal anti-hacking law because it was the user, with the agent’s help, who did so. The case was sent back to the federal district court in San Francisco.
Copyright suits brought by Dow Jones and The New York Times remain pending in federal court in Manhattan; Perplexity has asked the court to narrow the Times case, Reuters reported in March.
From Chennai to Berkeley
Mr. Srinivas, who studied at the Indian Institute of Technology Madras in Chennai before coming to Berkeley, has described choosing Berkeley because “when it came to AI, I think Berkeley was the most happening university at the time,” he told Berkeley Engineering in 2025. He interned at OpenAI after his first year, worked there for a year after finishing his doctorate, and co-founded Perplexity in 2022 with, among others, Andrew Konwinski, who also holds a Berkeley Ph.D. in computer science.
“Ph.D. students write papers, and for every sentence they write, they reference another paper,” Mr. Srinivas told the engineering school. “We baked that principle into an AI chatbot, and that ended up becoming Perplexity.”
Mr. Abbeel, his adviser, told the school that Mr. Srinivas “was the kind of student who not only brought great execution, but who was also capable of bringing great vision.”
The Batch API suggests where that vision is headed: less a single search box than a utility that other companies rent by the request. Perplexity’s press office did not respond to a request for comment on the launch, nor did David Bamman, a Berkeley School of Information professor who studies natural language processing, or the Berkeley Artificial Intelligence Research lab.
Marcus Thibodeaux covers UC Berkeley alumni in business, startups, venture capital, industry and government, from Sand Hill Road boardrooms to Sacramento and Washington, for the Inquirer’s Berkeley Alumni desk. More from Marcus Thibodeaux
