ChatGPT screenshots no longer fade part of the answer into white where the page's window ended.
Perplexity and Grok take include.screenshot: a full-page PNG of the answer's public page, the one include.html copies. Free.
Lower add-on prices
include.html on ChatGPT is free. Every ChatGPT include flag now costs nothing extra.
Google News pages after the first are free.
Google Search pages after the first, the AI Overview and each page with a search operator cost less. Engine prices are unchanged. See /pricing/.
Affiliate commission on Gemini and Perplexity
The affiliate commission now also accrues on what referred accounts spend on Gemini and Perplexity, next to ChatGPT and Google News. See /affiliates/.
Full Perplexity answers, faster Gemini, places and pictures
A Perplexity answer no longer draws on earlier questions asked through the same perplexity.ai account: each one is asked as perplexity.ai's incognito mode asks it, with no memory of what came before.
Perplexity answers are now the ones perplexity.ai writes for a signed-in user, with their sources and citationPills: its full answer whenever one is available (see answerDepth below), no longer a few sentences when one is not.
related_queries is no longer returned: perplexity.ai shows these users no follow-up questions. The field stays in the contract and is absent.
include.html links to a copy of the answer's perplexity.ai page.
Plain text answers on every engine read as the page shows them: & instead of &, no blanks at line ends, at most one empty line between paragraphs, and bold in Chinese, Japanese and Korean answers no longer leaves ** behind.
AI Mode answers now include the places, map pins and ads that Google shows next to them.
AI Mode answers no longer repeat each section heading: the text shows it once, as Google's page does.
Google Search and AI Mode return include.screenshot: a full-page PNG of each results page or of the AI Mode answer.
Screenshots no longer show a block drawn over the next one when the page was still animating.
Grok source previews hold the page's own text: no more date lines, menus, language pickers or styles at their start.
Links on every engine come without click tracking (utm_*, _ga, gclid, fbclid, srsltid and similar): the page's own address.
Source descriptions read as prose: a snippet cut from a table or list no longer carries its | separators, list stars or heading marks.
Perplexity sources carry the snippet Perplexity shows under each, as description.
Google Search: every organic date reads as Google shows it (Dec 30, 2025); displayedLink is always an address, never a view or follower count; peopleAreSaying returns again.
AI Mode nests list items with spaces, as the other engines do, and ChatGPT's text no longer ends a paragraph with a label whose citation chip was removed.
Copilot's and Grok's tables read with one space around each cell, Gemini's list items with one space after their mark, and no answer shows a doubled blank where a citation chip stood.
Source labels on every engine lose a stray blank inside their brackets, and a social post's title in Google's peopleAreSaying reads on one line.
New request field answerDepth: any (the default) returns Perplexity's full answer when one is available and its short one otherwise; full returns only the full answer or a retryable error that isn't charged; quick returns the short one. result.provenance.answerDepth says which came back. Other engines have one depth and ignore it. Same price for every value. See /docs/engine-features/#answer-depth.
Perplexity answers return the places on Perplexity's map as result.places (name, address, phone, rating, review count, coordinates, photos; hotels included) and the pictures shown with the answer as result.images.
ChatGPT answers return the local businesses they name as result.map (address, coordinates, categories, rating, review count, phone, website) and the pictures shown in the answer as result.images.
Perplexity source labels are the page title alone: a title no longer starts with its site's breadcrumb, such as docs.github.com › actions.
Gemini answers arrive in about 13 seconds instead of about 55: a request finishes once Gemini has shown the whole answer, and a warm page takes the next question. include.rawWire now holds the messages Gemini's page received for that answer, each base64-encoded, the detected location still removed.
Google Search returns the cards of Google's Recipes carousel as result.recipes, at no extra cost.
ChatGPT: with include.shopping, an entity ChatGPT links to a product carries entities[n].product, its price, URL and merchant offers.
Perplexity costs less: $1.50 per 1,000 requests on Starter down to $1.00 on Max. See the pricing page.
No per-account request rate limit: send as many requests a second as you need. 429 RATE_LIMIT_EXCEEDED now comes only when one address sends too many unknown API keys.
No fixed count of queued tasks: your queue holds about 100,000 tasks of a typical prompt at a time, more of shorter ones. A submission that does not fit is refused whole with 429 QUEUE_LIMIT_EXCEEDED, whose details say how many more fit.
X-RateLimit-Limit and X-RateLimit-Remaining are still sent and are deprecated: both are always 100000, meaning no limit.
SDKs 0.4.0: visibility.measure takes any number of samples (MAX_SAMPLES is gone), and clearQueue / clear_queue returns clearing: true while a long queue is still being deleted.
No hourly usage cap: however many requests an account runs, its tasks never wait for its past use.
An async task that has not started within 4 hours of creation fails with QUEUE_WAIT_EXCEEDED and is not charged; the bound was 72 hours.
DELETE /v1/async/queue answers within seconds however long the queue is: a queue too long to clear in one call answers 202 with clearing: true, and the rest is deleted without further calls.
Every engine takes prompts of up to 2,000 characters; Perplexity, Copilot, Grok and Google AI Mode took up to 10,000. A longer prompt is refused with a validation error that states its length and the limit.
New prices and spend per API key
New prices on every engine and add-on. Perplexity, Grok, Copilot, AI Mode and Google Search now cost the same on every tier. See the pricing page.
The dashboard's Usage page shows spend per API key, and the usage CSV names each row's key. Each task's detail page names its key.
Free page screenshots, free People also ask answers
include.screenshot returns a full-page PNG of the answer page for ChatGPT, Gemini, AI Mode and Google Search, at no charge. Other engines refuse it with a 400.
include.paaAioverview is free. Only include.aioverview is billed as the AI Overview add-on.
One ChatGPT price, and no charge for unfinished answers
ChatGPT's include.rawResponse and include.shopping are free: every ChatGPT request costs its base price, whatever it includes.
An answer the engine never finished is never returned: it is retried, and a request that still does not finish fails and is not charged.
The terms now state that you may keep results as long as you choose and show them to your own customers, a right that survives the end of service.
Prepaid USD balance replaces plans and subscriptions
Your account now holds a balance in US dollars. Top it up by any amount from $1 on the dashboard's Billing page, and save an amount to top up with one click there.
Every account gets 100 free requests on any engine, no card. Approved bonus programs (G2 review, backlink, Reddit intro) add 1,000 free requests once per program.
Price tiers keep their names (Starter to Max) and are reached by your spend over the last 30 days; bigger tiers cost less per request.
Plans, monthly subscriptions, billing cycles, annual billing and overage are gone. API field names stay the same; X-Credits-Charged and remaining now carry dollar amounts.
No per-plan concurrency cap
Plans no longer limit how many requests run at once. Send the volume you need; the API paces execution against each engine's rate limits and your one-second rate window, and work beyond that waits in your queue.
Annual billing's extra concurrency on Max is gone with it — annual terms keep the monthly price advantage.
Lower Google, Google News and AI Mode prices, whole-number add-ons
Google Search and Google News cost 34 credits per results page (was 35), the AI Overview 34 (was 35), and AI Mode 68 credits per task (was 70), on every plan.
Gemini costs 5.5 credits per task (was 5.6). ChatGPT stays at 7.
The synchronous surcharge and ChatGPT's rawResponse/shopping add-on are 7 credits (were 6.7); an AI Mode product cluster is 3 credits, at most 18 (was 3.3, at most 19.8).
Max billed yearly keeps 30% more concurrency; it no longer adds 5% more credits.
Free search-query and ads flags, product cards and cleaner Gemini answers
ChatGPT: include.searchQueries and include.ads are free. Both lists come back empty on the answers served today; the fields stay in the schema. The +6.7 add-on now covers include.rawResponse and include.shopping, charged once.
ChatGPT: brand and product names shown as chips in the app now stay in text and markdown, product cards come back in shoppingCards (title, price, rating, reviews and the store in offers[0].merchant_name), and tables come back as Markdown tables.
Gemini: comparison tables come back as Markdown tables, card markup no longer appears in text, and sources lists only the pages the answer cites, once each.
include.html is refused with a 400 on Gemini and Google News, before anything runs or is billed. On ChatGPT, snapshot links now open as soon as the result is delivered.
Country coverage per engine
ChatGPT answers from 19 countries. Hong Kong is not among them because ChatGPT is not offered there.
Gemini answers from 18 countries. A country is listed only where answers can be served from inside it.
Google News takes any country code.
GET /v1/countries?model=<engine> returns each engine's list, and a request for a country outside it is refused instead of being answered from elsewhere.