Parallel Web Systems announced the Parallel Decisions API on X on 27 October 2026, calling it “the first web-grounded decisions engine.” The company says the API is live at $0.001 per request, with 5,000 free requests to start.
Parallel’s launch post, dated 25 October 2026, describes it as “a decision engine with world knowledge built in.” Its pricing page lists the Decisions API at “$0.001 for up to 100 questions” per request.
Short explainer · Parallel Decisions API
A vertical news Short narrated from the published text of this piece and related FourWeekMBA pieces.
Business Pill · Usage-based pricing
An evergreen Business Engineer explainer on usage-based pricing: paying for work done rather than for seats. It explains the general idea and does not discuss Parallel or decision models.
The key insight: As we read it, Parallel is selling a web search and a decision in one call: its pricing page lists the Decisions API at $0.001 for up to 100 questions per request, the same floor price as its Search API, which returns 10 results.
What Parallel Launched
A decision model does not write prose. It answers a yes-or-no or multiple-choice question and returns a probability. Parallel’s September post describes Jev, a model from TypeSafe AI, this way: “it returns categories, scores, and probabilities instead of generating text.”
Parallel’s pitch is that those models only know their training data. In its X post, the company wrote: “Decision models like Jev can answer in an instant, but they lack the recency of current news. Now you can get both in a single call.”
The launch post gives an example: “A model like Jev can’t answer whether to lock your mortgage rate today, because it doesn’t know where rates closed yesterday.” It says the Decisions API “grounds every decision in real-time web context” and “returns citations with every answer, so every result can be audited.”

What Was Posted on X
Parallel Web Systems announced the Decisions API on X on 27 October. The post calls it the first web-grounded decisions engine, says decision models like Jev lack the recency of current news, and lists a price of $0.001 per request with 5,000 free requests, plus example uses such as fact-checking content, refreshing stale banking records and tracking competitor e-commerce inventory. This piece uses the post for the launch claim and the price.
@CompleteSkeptic wrote on X: “smart programmable routing with jev as a brain!”
Elvis Saravia highlighted on X a new paper comparing decision models with fine-tuned rerankers, which tested six Qwen backbones from 0.6B to 27B reranking the same BM25 top-100 candidates on ten collections, and advised comparing both on your own queries before replacing a reranker with a general decision model.
How Parallel Prices the Call
Parallel says one call can carry several questions about the same company, product or policy, and that they “all share a single call.” The launch post puts the price at “10x lower cost than a standalone model and just $1 per 1,000 calls.” It does not name the standalone model behind the 10x comparison.
Parallel’s pricing page lists Decisions API latency at 200ms to 1s, synchronous, with a rate limit of 600 requests per minute. On the same page, the Search API is listed at “$0.001 – $0.005 for 10 results” and the Task API at $0.005 to $2.4 per request.
By our arithmetic from that page, the Decisions API shares the lowest list price in Parallel’s line-up, $1 per 1,000 requests, with the Search and Extract APIs. The units differ: a Search request returns 10 results, and a Decisions request can hold up to 100 questions.
What Parallel Says It Tested


The launch post shows one request with four questions about Vercel, “answered from one retrieval of 10 sources in 681 ms for $0.001.” One of the four answers, the employee band, came back at confidence 0.55 and was “routed to review”, the post shows.
Parallel also describes two projects it built before launch. On bank checks, it wrote: “We rechecked 426 public companies against the live web with the Parallel Decisions API in under ten seconds for 85 cents, and 98% of its flags held up.”
On sales leads: “We ran 500 inbound leads through the Parallel Decisions API in under nine seconds for about 50 cents, and 97% of the leads it routed on its own landed in the right segment.” The company adds: “We will use Decisions to power our own contact.parallel.ai.”
The X post lists other uses the company has tried, including fact-checking content, refreshing stale banking records and tracking competitor e-commerce inventory. The launch post says the API is also a free tool in the Parallel MCP server, called web_decisions.
Who Is Behind Parallel
TechCrunch describes Parallel Web Systems as “the AI agent-tool startup founded by former Twitter CEO Parag Agrawal.” In April, Parallel announced “our $100 million Series B at a $2 billion valuation,” led by Sequoia Capital.
The company said the round “more than doubles our valuation from five months ago and brings total capital raised to $230 million.” The same post said: “More than 100,000 developers are building on Parallel today.”
A Crowded Market for Decision Models
Parallel’s own October post charted Jev’s adoption: “Jev 1.13 reaches 3.27B calls, GPT-6 Luna 845M and MiMo-V2.6-Flash 365M,” in cumulative decision calls since release. The same post said Jev’s hosted API “offers the best combination of cost and latency among the models we tested.”
The decision models we have covered are priced per million tokens, while Parallel prices the Decisions API per request, so the list prices cannot be compared directly. Parallel’s product adds a web search step before the answer.
Connecting the Coverage
On 22 October we reported that OpenAI opens Decisions API at 2.4x the price of rival Jev. On 24 October we reported that Perplexity’s decision model runs free on a Mac, no API key.
In September we covered Parallel’s Parag Agrawal on agents, advertising, and a 1000x design premise. And earlier today we reported that Nace says its Drex search cut coding agents’ steps by 62%, another bet on search built for agents rather than people.
The Structural Read
As we read it, the product targets a gap Parallel names itself: decision models such as Jev answer from training data, and Parallel says its API grounds every decision in real-time web context and returns citations.
As we read it, batching is the pricing lever: Parallel says several questions about one company or product share a single call, and its Vercel example shows four answers from one retrieval for $0.001.
As we read it, the evidence so far is Parallel’s own: the 426-company and 500-lead runs, with 98% and 97% figures, come from projects the company built before launch.
Parallel, Decisions API launch post, 25 October 2026
“A model like Jev can’t answer whether to lock your mortgage rate today, because it doesn’t know where rates closed yesterday.”
Three Implications
FOR DEVELOPERS Parallel says the API is live with 5,000 free requests, and is also a free tool called web_decisions in its MCP server.
FOR BANKS AND SALES TEAMS Parallel says it rechecked 426 public companies for 85 cents and routed 500 inbound leads for about 50 cents in its own pre-launch projects.
FOR DECISION-MODEL VENDORS Parallel’s own October post said Jev 1.13 had reached 3.27B cumulative decision calls; the Decisions API adds live web context to the same kind of yes-or-no answer.
The Business Engineer Lens
This story maps onto the Business Engineer framework Tokenomics: The Economics of AI.
The framework says of the token: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
As we read it, Parallel prices the Decisions API on a different unit: per request, of up to 100 questions, according to its pricing page, rather than per token. The framework was written in May 2026 and does not discuss Parallel or decision models.
What Is Not Established
The 98% and 97% figures come from Parallel’s own pre-launch projects; we have seen no independent test of the Decisions API. Parallel does not name the standalone model behind its 10x cost claim.
The launch materials give no usage, customer or revenue figures for the Decisions API. The launch post is dated 25 October, while the X announcement went out on 27 October; the documents we read do not explain the gap.
The Bottom Line
Parallel Web Systems says its Decisions API answers yes-or-no and multiple-choice questions from live web sources, with citations, at $0.001 per request of up to 100 questions, and says its own tests rechecked 426 companies for 85 cents with 98% of flags holding up.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Launch details are from Parallel Web Systems’ own post on X (27 October 2026), its Decisions API launch post (dated 25 October 2026), its pricing page and its earlier blog posts, read in full on 27 October 2026. The accuracy and cost figures for the 426-company and 500-lead runs are Parallel’s own tests. Funding details are from Parallel’s April 2026 announcement and TechCrunch. Figures labelled “by our arithmetic” are ours. This is not investment advice.
Sources: Parallel Web Systems on X, Decisions API launch (27 Oct 2026) · Parallel, Introducing the Parallel Decisions API (25 Oct 2026) · Parallel, pricing page (read 27 Oct 2026) · Parallel, Series B announcement (29 Apr 2026) · TechCrunch, Parallel Web Systems hits $2B valuation (29 Apr 2026) · Parallel, Testing out Jev (18 Sep 2026) · Parallel, Are all System One models created equal? (14 Oct 2026)









