Genspark

The Membership Plans page gives Free 100 credits a day, 1 GB of AI Drive storage, limited access to models and agents, and 4,000 words a week of Speakly voice input. The refill is not permanent: once an account has used its lifetime allowance of free credits, the daily top-up stops and only the remaining balance keeps working. The page does not say how large that allowance is.

On Free, AI Chat spends those credits too. The AI Chat help page names Claude Haiku 4.5 as the default model and says chat is limited by the credit balance rather than by the model picked. Attached files can be up to 500 MB on Free and up to 1 GB on paid plans.

Paid plans run on three meters. Monthly credits start at 10,000 on Plus and 125,000 on Pro. A weekly Free Quota covers text and image tasks in Standard mode across Super Agent, AI Slides, AI Sheets, AI Docs, Deep Research, Fact Check and SecondBrain without spending credits; its 7-day window starts with your first message, and Pro gets 10 times the Plus balance. When the quota runs out, tasks switch to credits unless Ask First is turned on. AI Chat has its own fair-use cap over a fixed 5-hour window.

The credits guide ties cost to model tier, context length, image resolution, video and audio length and the number of versions generated. It also lists what multiplies a bill:

• A regeneration costs the same as the original run.
• Failed agent retries are charged.
• Every message in a long chat reprocesses the full context, so later replies cost more.
• Auto-select in AI Audio and AI Music generates several versions and charges for each.

For actual prices the guide sends users to the live pricing inside Genspark, so no public page states what a Deep Research run costs.

With Search Web switched on, AI Chat searches in real time and links the pages behind its answer. The Fact Checker page describes a stricter chain: the agent runs up to 30 rounds across Google Search, Google Scholar, academic databases and news sources, captures screenshots of evidence pages, and has a second model confirm that each screenshot supports or contradicts the claim. Each claim then gets a TRUE, FALSE or PARTIALLY TRUE verdict with source URLs and verbatim quotes.

The Terms of Service still say users cannot rely on content the service generates and carry the risk of its accuracy, so open the quoted page before reusing a verdict. If the main need is a search engine that answers each question with linked sources rather than an agent that builds files, the Perplexity profile covers that case.

The Privacy Policy names OpenAI, Anthropic, Google, xAI and ElevenLabs as the main AI providers that receive inputs, stores data on Microsoft Azure, and deletes account data within 30 days of an account being closed. The terms give Genspark a licence to submitted content for operating, promoting and improving the service and for developing new ones.

The Team and Enterprise page says both plans are automatically opted out of model training and holds SOC 2 Type II and ISO 27001 badges. The pages read make no equivalent statement for Free, Plus or Pro, which matters before uploading unpublished research or client documents on an individual plan.

An upgrade applies at once with a prorated charge, and the difference in monthly credits is added straight away. A downgrade waits for the next renewal, and unused credits from the higher plan are lost. Cancelling keeps the plan to the end of the paid period; the account then returns to Free, where daily credits flow only until the lifetime allowance is used. Annual billing charges the year up front but releases credits monthly, so a heavy research month cannot borrow from the next one.

Subscribers who started before September 18, 2026 keep unlimited zero-credit AI Chat and AI Image use through December 31, 2026 if the subscription stays active. From January 1, 2027 every Plus and Pro member moves to the core-model cap, after which chat draws on monthly credits.

1. Create a free account, open the Credits page and note the balance.
2. Ask Super Agent one research question you can judge, such as a regulation or study you know well, and note the credits used once the usage log catches up, which can take a few hours.
3. Run Fact Check on three specific claims from the result and open every quoted source page.
4. Mark any verdict whose quote does not appear on the linked page, or whose page says something different.
5. Divide 10,000 by the credits one run used to see how many such runs a month of Plus would cover, counting retries as full runs.

Accept if the quotes match their pages and one run leaves room for your monthly volume. Reject if verdicts rest on quotes you cannot find, or one question uses a large share of the free balance. No result is recorded here.