The page documents material changes to Perplexity Sonar API citations: Citations now arrive as an output item of type search_results in the Agent API, with a queries array and a results array containing id, url, title, snippet, source, date and last_updated; the Agent API endpoint is POST and Sonar Chat Completions…
From Perplexity's pricing page (November 8, 2026): Agent API is billed per token at the model's rate plus tool fees; exact costs depend on model/effort level (fast, low, medium, high, xhigh, ultra) with a new Ultra level introduced November 5, 2026; Search API $5.00 per 1,000 requests (or $1.00 with Fast Search);…
The page aggregates pricing and changelog details from multiple providers as of 2026-10-31 to 2026-11-06, but it does not publish a single, verifiable material change to an individual system. It references many price changes and feature notes across providers, yet no explicit, unified new change to a specific system…
URL: Exa revenue $100M annualized, run rate 22x multiple; DeepSeek API floor $0.75 per 1,000 requests; Exa Lite $4 per 1,000; other prices listed (Brave Search $5, Perplexity Sonar $5, Exa Auto $7, Tavily basic $8). Data sources: Exa, DeepSeek, Yahoo Finance, DEV Community independent test; date references include…
URL fetching for web grounding added with fetch_url pricing $0.001 per successful fetch; Agent API effort levels updated with new model pins and step limits (fast uses openai/gpt-5.6-luna fallback, high uses openai/gpt-6.1-terra with 40 steps, xhigh uses openai/gpt-6.1-sol with 300 steps; max_steps/levels…
Improvements to every Agent API effort level.
and referenced as Free, no sign-up and Sign in for Research with higher limits; Parallel MCP 2.0 combines web search (web_search), web_fetch, web_decisions, web_research with upcoming tools (Monitors, Data Connectors) and enterprise configurability; new MCP 2.0 launched November 5, 2026.
documents that the OpenAI API web search guide now recommends gpt-6-astra for API web search builds and multi-source research, lists o4-mini as shut down, and removes gpt-4o and older workflows; context limit remains 128k; two workflows replace three search types.
Perplexity API changelog updates: URL fetching costs $0.001 per success; supports setting max staleness; high and xhigh effort levels support more steps and higher token limits; ultra-preview added; undocumented ultra name removed; max steps increased to 500; Python library and TS SDK lack mode parameter for web…
The page announces that Perplexity’s Decisions API can now search the web before it answers by adding a setting "mode": "web_grounding" to any Decisions API request, enabling Web Grounding with Fast Search and evidence in the response; benchmarks (PWS-Decisions) show improved accuracy (from 51% to 86%) and latency…
URL: The page documents various updates to SearchApi including October 2026: Update with new APIs and features, MCP Tool Bundles updates, ChatGPT API features, and MCP: OAuth 2.1 and the Web Search Bundle. No concrete pricing changes or end-of-life noted on the excerpts provided.
The page documents material changes to Perplexity and related grounded AI tooling: direct URL fetching at $0.001 per successful fetch, new agent effort tiers including ultra-preview with higher step limits and token caps, and API change renaming preset to effort (backwards compatible). It references Perplexity API…
Not applicable (page is a log of changes; no single model IDs, tiers, prices, limits, and dates to extract as a standalone material claim for a specific change).
The page documents a material, newly published change to web-search/deep-research pricing and metering models across Perplexity Sonar Deep Research, OpenAI deep research models (o3-deep-research and o4-mini-deep-research), and Google Deep Research agents (standard and Max), including published October 2026 rates,…
The page documents updated Perplexity API pricing as of November 2, 2026, detailing Sonar, Agent API, and Search API token and tool invocation fees, with Sonar Chat Completions ended Sept 27, 2026 and Sonar now migrated to Agent API pricing structure (token rates plus per-tool fees).
Perplexity: November 2026 (Nov 2, 2026).
DeepSeek Search: $0.75 per 1,000 successful requests. DeepSeek Fetch: $0.75 per 1,000. This change took effect 00:00 Beijing time on Nov 1, 2026 (16:00 UTC Oct 31). Previously, server-side calls were free through Oct 31; as of Nov 1 they are billed at the same rate as direct API calls. Pricing applies to Search and…
I benchmarked 4 web search APIs for AI agents. Here's what each one is good at. - DEV Community.
The page announces Expand Search, a web search API for agents, priced at $1.50 per 1,000 searches, as part of Expand's products following Expand Fetch.
URL not provided in the extracted extras; page discusses price drop from 5 credits to 2 credits per call for browserfoundry deep search (Startup: $1.78 per 1,000 requests; Essential: $5.33) and 96.5% SimpleQA coverage.
Source: Artificial Analysis, data as of 19 October 2026. The page presents a benchmark of 26 search products from 14 providers (12 in public table) showing index scores and costs; Exa Search (auto) and Perplexity Search (medium) lead with an index of 80 and costs/time detailed in a public table. No explicit model IDs…
Introducing Monitors v2 | Exa Blog.
Perplexity: October 2026 (Oct 29, 2026).
(FinSearchComp, T3) shows Parallel most accurate at up to 2x lower cost than Exa and Perplexity; FrontierFinance benchmark shows Parallel ties for top score at 38% lower cost than Exa and beats OpenAI, Perplexity, and Tavily on both accuracy and cost; private PWS-Finance benchmark shows Parallel beating OpenAI's…
The page documents multiple search/retrieval APIs (Exa, Parallel, Tavily, Brave Search API, Perplexity’s Search and Agent APIs, Firecrawl, Goggles, etc.) with details on latency, outputs, and billing units (e.g., Parallel: 200ms–3s synchronous, 600 requests/min; Tavily: 1 or 2 API credits per request; Perplexity’s…
Okara published an independent study of 230 B2B software-buying answers from web-enabled ChatGPT, Claude, Gemini, and Perplexity (including Claude Sonnet 5.5, Gemini 3.5 Flash, and Perplexity Sonar Pro). It reports that systems shared an average of 6% of cited domains for the same question, despite agreeing on…
Parallel announced that its Decisions API is available through OpenRouter, allowing developers to use an existing OpenRouter API key without a separate account. The model ID is `parallel/decisions-fast:free`, and Parallel offered a seven-day free trial. The web-grounded API returns structured answers with…
Google deprecated `gemini-embedding-2-preview` effective November 6, 2026, and began routing API requests for it to `gemini-embedding-2` at the latter model’s rates. Google says the replacement uses the same API schema, pricing, and embedding space, so no other code changes or vector re-indexing are needed. This…
Arca published a research-tool leaderboard based on 50 life-science questions from the University of Pennsylvania’s ResearchQA benchmark. Arca reported a score of 84.50/100, followed by Gemini 3.1 Pro with web search at 83.53 and GPT-6 Astra with web search at 83.14; Consensus, SciSpace, and Elicit scored lower in…
Parallel launched MCP 2.0, a hosted MCP server offering web_search, web_fetch, web_decisions, and web_research. The first three are available without an API key at lower rate limits; web_research requires authentication and is billed at the underlying API rates. The release also adds controls for tool selection,…
Perplexity added optional live web grounding to its Decisions API: setting mode to web_grounding runs Fast Search for each question and supplies retrieved evidence to the decision model. Perplexity reports that accuracy on the 100-question PWS-Decisions public set rose from 51% to 86%, with median response time under…
DeepSeek began charging for Search and Fetch API calls at $0.75 per 1,000 successful requests for each API, including server-side tool calls through compatible coding-agent environments. The billing change took effect at 00:00 Beijing time on November 1, 2026.
SerpApi fixed its Google Search API handling of news results that were returning goto links. This is a concrete reliability correction to search-result retrieval.
Snowflake announced that Deep Research in Snowflake Intelligence can now research the web as part of an investigation when web search is enabled. Web findings are cited alongside findings from the user's data, while the web research itself does not access that data.
Google made the Gemini API Computer Use tool generally available across the Interactions API and GenerateContent API. The GA release makes the tool available for agents to interact with computer interfaces, a capability relevant to autonomous web research workflows.
Parallel launched its Slack agent, which monitors the web for specified changes, posts alerts with sources in Slack channels, and answers follow-up questions in threads. This adds a new Slack-based surface for Parallel’s web monitoring and research capabilities.
Parallel launched Parallel for Finance, an index-and-ranking upgrade for financial-agent research available across its APIs without a new plan or subscription. In its published evaluations, Parallel Advanced scored 57.4% on FinSearchComp T3 at $161.10 per 1,000 questions, compared with Exa Auto at 51.4% and $226.20…
Google released the Gemini API CLI in beta, a stateless command-line client for agents, scripts, and CI that calls the Interactions API and can run Gemini managed agents, including Deep Research. It supports background research jobs and machine-readable output, adding a new interface for integrating Gemini research…
Parallel launched the Decisions API, a live-web-grounded decision service that returns typed answers with citations and probabilities. It supports batching multiple questions against shared context; Parallel says one retrieval can answer up to 100 questions for $0.001 per request. The API is intended for fast…
Google’s Gemini API release notes say that, starting October 26, new paid-tier projects using the Generate Content API store requests by default (store=true). Existing projects keep their current setting; developers can opt out per request or disable logging in the Logs and Datasets settings. This is a material…
Parallel published a web-search API benchmark report describing its agent setup and results. It reports that Parallel Advanced scored 97% on SimpleQA Verified versus 95% for Perplexity, and 57.6 on WideSearch versus 55.9 for Exa and Tavily. These are Parallel’s reported benchmark results, and the page explains the…
Parallel announced EU data residency for its Search and Extract APIs. Customers can direct requests to eu.parallel.ai instead of api.parallel.ai; the company says the EU endpoint supports a foundational set of Search and Extract features, with more planned. This adds a regional API endpoint for customers requiring EU…
Exa published an evaluation comparing the reranker used in its production search against leading standalone reranking APIs, with four additional evaluations covering navigational intent, freshness, authority under adversarial content, and coding. In its navigational evaluation, Exa reports macro nDCG@10 of 0.564 for…
Exa announced Exa 3.0, describing it as its largest search-quality improvement and claiming it leads on eight tested benchmarks, including BrowseComp (91.8%) and WideSearch (56.1%). The release also introduces Exa Lite at $4 per 1,000 requests, makes Dynamic Highlights generally available at no additional charge…
Parallel published an updated Search Capability Leaderboard evaluating how models use web search and synthesize results. The October 17 update reports Claude Opus 5.5 as the Search Intelligence leader with a score of 75.4, MiMo v2.6 Pro as the most cost-efficient model at $76.5 per 1,000 tasks, and an average search…
Parallel became the default search provider for Browserbase Search API requests and Browserbase Agents. Browserbase users can select a provider per request, and the integration supports Parallel’s fast and advanced search modes.
Keenable announced Dynamic Snippets for agentic search, claiming the feature uses 50% fewer tokens. This is a material retrieval-efficiency capability claim; the gathered information does not provide further details on its implementation or measurement methodology.
Parallel published a comparison of hosted and open-weight System One decision models on search-result reranking, page-change classification, low-quality content filtering, and query routing. The report says prompt tuning raised Rune 26B’s page-change-classification AUPRC from 79.8 to 89.1 on the calibration set and…
Parallel published an evaluation of its Task API on 288 LABBench2 biology-research questions. It compared runs using the same agent and model with public-web access alone against runs with additional Parallel connectors to PubMed and ClinicalTrials.gov. The report describes using web search for open-ended retrieval…
Changelog 2026-08-24: citations and confidence scores are now emitted per list element (separate FieldBasis entries) instead of once per field, without the beta header.
The same GPT-6 Astra agent with Exa Search + Contents or DeepSeek Search + Fetch. Compare reviewed accuracy, completion time, and itemized costs on LiveBrowseComp and EvoBrowseComp.
One GPT-6 Astra agent, five web tool stacks. Exa and built-in web tie at 11/26 on LiveBrowseComp, with Keenable one answer behind. Explore accuracy, run duration, and every public task verdict.
Claude Code vs Codex CLI under three search backends (builtin, Exa MCP, Valyu MCP) on a frozen 30-task LiveBrowseComp subset. claude-exa wins at 60%; the search backend moves accuracy up to 33 points, the agent at most 7.
Five deep-research APIs plus a Perplexity calibration anchor on a frozen 30-task ResearcherBench subset. Coverage is a three-way race; faithfulness splits the field in two; parallel has the best overall profile.
Nine deep research APIs on a frozen 20-task subset of DeepResearch-Bench-II. Exa effort tiers against each other, then one flagship tier per provider.
