Perplexity Search API vs Parallel in 2026

Run the same question through both formats

Each API keeps its own request and its own JSON on SpecStream. This script sends one question in Perplexity's format and in Parallel's, with one key, and prints the result count, the seconds each call took and the first link.

import os
import time
import requests

KEY = os.environ["SPECSTREAM_API_KEY"]
BASE = "https://api.specstream.ai"
Q = "who maintains the requests library for Python"

calls = {
    "Perplexity": lambda: requests.post(f"{BASE}/perplexity/search", headers={"Authorization": f"Bearer {KEY}"},
                                        json={"query": Q, "max_results": 10}, timeout=30).json()["results"],
    "Parallel": lambda: requests.post(f"{BASE}/parallel/v1/search", headers={"x-api-key": KEY},
                                      json={"search_queries": [Q], "mode": "fast"}, timeout=30).json()["results"],
}

for name, call in calls.items():
    start = time.perf_counter()
    results = call()
    seconds = time.perf_counter() - start
    print(f"{name}: {len(results)} results in {seconds:.2f} s, first: {results[0]['url']}")

The two calls cost $0.002 together, from the $10 of credit a new account gets. Change Q to a question from your own logs and run it again.

What 10,000 searches a month cost on each

Take an agent that makes 10,000 searches a month of 10 results each, at list prices.

SpecStream: $10 on either path. $50 if every Parallel call runs advanced.

Perplexity: $50 on standard web search, $10 with Fast Search.

Parallel: $10 on turbo or fast, $50 on basic or advanced.

At 20 results a call, Parallel adds $1 per 1,000 results beyond the first 10, so fast costs $110. SpecStream costs $22. Perplexity keeps one price per request whatever the count: $50, or $10 with Fast Search.

The most accurate web search API in the world

The SpecStream Search API scores a 97.9 F-score on all 1,000 official SimpleQA Verified questions, with 978 correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks, and the Extract API scores 0.908 F1 on WCXB. The scores, the method and the other APIs' published scores are at specstream.ai/benchmarks.

The Perplexity and Parallel figures in the table come from Parallel's own run of September 2026, a GPT-5.6 Sol agent on 100-question samples. Ours are on every official question.

What each path reads and returns

Perplexity: POST /perplexity/search reads query, max_results up to 20 (50 for people), search_type, search_domain_filter with a - before a domain to leave it out, search_recency_filter, search_after_date_filter and search_before_date_filter, max_tokens and max_tokens_per_page. It returns results[] with title, url, snippet, date and last_updated. Send one query a call: a query array of two or more is refused with a 422, so a fan out of 5 queries is 5 calls, $0.005 together. country, search_language_filter and search_context_size are accepted and do not change the search.

Parallel: POST /parallel/v1/search reads search_queries (two or more are searched together), mode, advanced_settings.max_results up to 100, source_policy and excerpt_settings, and returns results[] with url, title, publish_date and excerpts. The search runs on search_queries; objective and fetch_policy are accepted and do not change it.

The official SDKs need only the address. Every parameter and reply field, with what we measured, is in the Perplexity and Parallel migration guides.

import os
from perplexity import Perplexity
from parallel import Parallel

key = os.environ["SPECSTREAM_API_KEY"]
perplexity = Perplexity(api_key=key, base_url="https://api.specstream.ai/perplexity")
parallel = Parallel(api_key=key, base_url="https://api.specstream.ai/parallel")

Zero retention on any key

Set zero_retention when you create a key, or later. A search made with that key keeps no request text and no answer after the reply. The charge keeps the account, the key, the request ID, the price and the time. Set eu_egress_only and the key's searches leave only from our server in Madrid.

Search, pages and rows on the same key

The same key reads up to 20 pages a call with the Extract API at $0.80 per 1,000 pages, a page with no text not charged, and runs Deep Search, which returns every match on the live web as structured rows at 5¢, 15¢ or $1 a row. The most a Deep Search run can cost is approved before it starts.

Exa

Search by type: $4 per 1,000 requests on lite and $7 on auto and fast. SpecStream answers Exa's POST /search at https://api.specstream.ai/exa, at $0.001 a search of up to 10 results.

Tavily

Search sold by credit: an advanced search is 2 credits, at $0.008 a credit pay as you go. SpecStream answers Tavily's POST /search at https://api.specstream.ai/tavily, in Tavily's JSON, at $0.001 a search.

Brave Search API

Web search at $5 per 1,000 requests. SpecStream answers Brave's GET /res/v1/web/search at https://api.specstream.ai, at $0.001 a search.

Serper

Google results as prepaid credits, from $1.00 per 1,000 on the 50k pack. SpecStream answers Serper's POST /search at https://api.specstream.ai/serper, and SpecStream credit never expires.