Count three things separately: your application or model plan, research-provider charges, and local taxes or currency conversion. Free retrieval still uses your assistant’s model allowance.
These are dated examples to help you compare options, not a promise of current prices or availability in your account. Prices and availability may have changed since these checks.
Start with what you already have#
If your current application supports the tools you need, try one useful task before adding subscriptions. The ChatGPT + TinyFish walkthrough records the actual calls and what billing information was unavailable.
Provider allowances and access#
Verification dates: Exa’s credit terms were checked on 26 September 2026; TinyFish’s Search/Fetch terms were checked again for the beginner walkthrough on 27 September 2026. Other provider examples below retain their 19 September 2026 source check.
Both Exa and TinyFish offer a way to get started without paying for retrieval immediately. Exa provides a credit allowance; TinyFish offers free Search and Fetch within rate limits. Choose based on the work you want to do and the connection your application supports.
Exa: explore search with a free credit allowance. Its official pricing documentation (new tab), checked 26 September 2026, describes $10 in free credits on signup, with the free balance resetting to $10 on the first of each month. Completing dashboard onboarding earns the first team a one-time $10 bonus. No payment method is required for the free tier. Usage consumes credits at the applicable rates, so how far the allowance goes depends on the operations you choose. Follow the official Exa setup guide (new tab) to connect it, then try a question within your free allowance.
TinyFish: start here if avoiding extra retrieval costs is your priority. Its Search and Fetch services are free within published limits: up to 30 Search requests per minute and 150 fetched URLs per minute on the wallet-based tier, even with a zero balance. The announcement says no credit card is required to start. This free offer covers Search and Fetch; do not assume it includes browser automation or every other TinyFish service. Official free-tier terms and setup options (new tab).
Create an account, then use a supported integration to connect TinyFish to your application. Its setup page includes the hosted MCP route. For a first task, ask your assistant to search for sources and fetch the relevant pages using only TinyFish Search and Fetch.
Octen: explore Broad Search when your question has several angles. Broad Search takes one question, breaks it into related subqueries, and searches them concurrently. That makes it an option for comparisons and surveys—for example, researching a tool's features, limitations, integrations, and pricing together. Octen also offers focused Search and page extraction. Follow the official Octen setup guide (new tab), which includes hosted MCP instructions for several applications. Check current usage charges before enabling it.
Perplexity is another option to explore. See its official setup guide (new tab) if its available capabilities suit your work. There is no need to connect all four providers before starting.
Regional application examples#
Choosing these parts separately also lets you explore combinations that fit your budget. This can be especially useful in India, Pakistan, and other emerging markets, or anywhere you want to get more from what you already spend. A coding-oriented application can serve as a workspace for research and everyday assistance when it supports the necessary skills and tools.
For readers in India, Cursor Start is a useful example. Cursor announced the plan on 28 July 2026 (new tab). Its launch blog (new tab) lists ₹649/month, including tax, billed in INR with UPI or card, along with more agent requests than the free plan and support for plugins, MCP servers, hooks, and skills. Those connections are what make it relevant here: you can explore pairing Cursor's included model access with a separate research provider.
This is an India-specific offer, not a price to assume is available in Pakistan or other markets. Model names and allowances can change; confirm current eligibility, included usage, and pricing in your Cursor account before subscribing. The price above comes from the launch blog; the public pricing page retrieved for this guide did not display the regional Start offer.
For a broader lower-cost option, Command Code lists its Go plan at US$1/month (new tab) and its GOAT plan at US$10/month (new tab). These are starting subscription prices with their own usage allowances, not promises of unlimited model use. Its payment documentation (new tab) also describes UPI for INR-billed plans and Alipay for USD-billed plans. Check the payment options available to you at checkout.
Two combinations to explore are Command Code Go + the multi-provider-research skill + Exa within its free credit allowance, or the same application and skill + TinyFish Search and Fetch within its free limits. Both pair a low-cost model plan with a way to start retrieving sources without additional provider charges. These are suggested combinations, not end-to-end setups tested for this guide. Confirm that the application exposes the connected tools to your chosen model.
If you already pay for a compatible application, start by exploring its supported connections. Account for three things separately: your application or model plan, any research-provider charges, and local taxes or currency conversion. Free retrieval still uses your assistant's model allowance when it reads sources and writes an answer.
Regional application examples: sources checked on 19 September 2026. Cursor Start details are attributed to its launch announcement; other listed prices and provider terms reflect the official pages retrieved on that date. Check the linked pages and checkout for changes before signing up.
Compare costs without guessing#
The beginner comparison used three searches and two fetch batches covering six URLs. It did not return billing metadata. The advanced Exa case study reports one run’s cost, not an expected cost for every task. Use measured call counts and clearly attributed provider terms; do not treat either example as a typical-task price.