Explore research inquiries with cited reports and reasoning traces. Choose between web research modes and developer APIs with different execution and pricing rules.
Used this tool? Rate it
Used this tool? Rate it
Apodex is an AI research service for questions that require gathering and reasoning over information. Its inquiry interface presents step-level reasoning traces and citations in reports. The provider describes these as verified briefs, but its consumer terms say outputs should not be treated as a sole source of truth.
Apodex US, Inc. is the contracting company named in the consumer terms for the hosted apodex.ai application.
A signed-in inquiry can be saved and revisited. The interface lists the ability to branch from a report and search across threads.
The provider separates the work into research, deeper reasoning and a discovery preview. The descriptions below are the provider's account of each mode, rather than a promise that a particular investigation or verification will succeed.
The research API accepts messages and a selected model through an OpenAI-compatible chat-completions interface. The provider documents API-key authentication and gives a model identifier for each research tier.
Research chat completions default to streaming; the core-model interface defaults to a non-streaming response. Integrations switching between the two must account for that difference rather than assuming that the same client defaults apply. Research requests may run agent steps and tool calls, while core-model requests return direct model output without the hosted research loop.
The documentation also describes a different error surface: core models return HTTP errors, whereas a deep-research failure may be carried inside a response body or stream with HTTP 200. This distinction affects how an application detects a failed task; checking only for a successful transport status does not interpret the research response body.
Apodex presents Deep Research as available for market overviews and background research. The provider also supplies example inquiries on the application page, including a macroeconomic question about interest rates and questions about medical evidence. Those are examples of research topics, not examples of independently verified outcomes from the service.
A background inquiry can bring several web sources into a single structured report. The documented Deep Research workflow searches, extracts and synthesizes material, and the application says citations are included.
Deep Solve is described for technical questions involving conflicting or sparse evidence. Deep Discover is described as a multi-agent workflow that explores alternative paths and uses designated verification roles, but its API is a preview requiring access.
The consumer terms set a minimum age of 18, or the higher age required to consent in the user's location. The US Apple listing classifies the app 4+, a platform rating that does not change the terms' age requirement. Business and API use also has a separate contractual scope from individual consumer access.
The hosted inquiry application is available on the web and describes sign-in, saved inquiries and report branches. US listings identify Apodex apps for iPhone and iPad, and Android devices. The consumer terms expressly cover the hosted apodex.ai application for individuals. The public subscription page requires sign-in to view a specific account's plan, balance, credit transactions and invoices.
The developer platform documents a programmatic API with model names, message input and API-key authentication. The core-model documentation identifies an OpenAI-compatible interface for direct model inference and describes a separate Anthropic-compatible protocol for the core models. Protocol compatibility is an integration format, not a statement that another company's models operate inside this service.
Consumer terms also restrict automated access except via an API key or where explicitly permitted. The core-model API is text-only: image, audio and file content parts are rejected. That restriction belongs to the documented core-model endpoint and should not be generalized to every Apodex research workflow. The non-streaming core-model contract also rejects a max_tokens value above 32,768; longer generations require the streaming path described in the documentation.
The consumer web product uses credits, while the API documentation lists token and tool-call prices. The web pricing page says credit consumption is dynamically calculated based on model token consumption and tool invocation costs. It does not state a fixed number of credits for every question on the inspected page; the actual amount depends on the request. The subscription page asks users to sign in to see their own plan, balance and credit transactions. In the November 12, 2026 rendered pay-as-you-go display, the page offers 300 free credits to start; checkout was not tested.
The public API price page lists the following rates as of November 12, 2026. The dollar amounts are USD per one million tokens, not web-app credit prices. Access labels are those shown in the documentation, and Deep Discover is marked preview.
| API tier | Input per 1M tokens | Output per 1M tokens | Access shown |
|---|---|---|---|
| Deep Research 1.1 | USD 5 | USD 20 | Available to all |
| Deep Solve 1.1 | USD 5 | USD 25 | Available to all |
| Deep Discover 1.1 | USD 10 | USD 100 | Preview, request access |
Input includes the initial request, including history and files as described by the API price page. Output includes the tokens produced while the research agent works, including reasoning and intermediate steps, rather than just the finished report. Hosted web search, web fetch and code sandbox calls are billed on top of tokens: respectively USD 5, USD 6 and USD 90 per 1,000 calls in the published list. These are separate usage components, so the length of the final report alone is not the metering basis.
The core-model API has a different price structure. Its list distinguishes input, cached input and output tokens; tokens already generated are billed even if the client disconnects. The listed 1.1 core rates per million tokens are USD 0.60, 0.06 and 3.60 respectively; the Mini rates are USD 0.20, 0.02 and 1.20. Core-model requests with more than 200K input tokens are billed at twice the listed rates for the whole request, including input, cached input and output.
For institutional accounts, the provider describes centrally allocated shared monthly credits that reset without rollover; purchased top-ups are described as remaining until used. Organization creation and deletion are not self-service in the institutional guide. Program applications request a project, team credentials and compute requirements before technical review and onboarding if approved. The program advertises a three-hour response SLA for technical inquiries across US and Singapore time zones.
The consumer terms say a subscription renews for successive periods unless it is cancelled. To avoid the next period's charge, the terms require cancellation at least 24 hours before the current period ends; cancellation takes effect at the end of the paid period, with access retained until then. App-store subscriptions have their own cancellation route through the relevant store.
The terms generally make paid fees nonrefundable, subject to stated exceptions and applicable law. Approved consumer refunds are typically returned to the original payment method within 5–10 business days. If the provider terminates paid services for a reason other than the user's breach, a policy violation or unlawful conduct, the terms describe a prorated refund for the unused prepaid period. The handling of a purchase through an app store is subject to that store's terms and process.
The service-credit terms distinguish promotional credits from credits purchased for money. Promotional credits expire according to the period stated at issuance; prepaid credits do not expire except to the extent permitted by applicable law. Credits are subject to the service-credit terms and are distinct from a recurring subscription charge.
ChatGPT deep research is another research workflow: its provider describes a research agent that scans sources and produces a structured report with citations. Its documented flow also lets users choose sources and interrupt the work to clarify or narrow the question. Apodex's documented research API emphasizes selecting a research tier, while its core-model API is direct inference without a hosted search loop. Elicit describes automating screening and data extraction for systematic reviews, while Wolfram|Alpha describes computed answers from its knowledgebase.
The provider's consumer terms require users to evaluate outputs for accuracy and appropriateness for their use case, including human review where appropriate. They also say output should not be the sole source of truth or a substitute for professional advice.
The consumer terms prohibit using an output about a person for a decision with a legal or material impact on that person, such as a credit, employment, housing, insurance, legal, educational or medical decision.
An independent analysis of the model family cautions that agent-team benchmark results and single-pass core-model evaluations use different setups. It should not be read as an independent hands-on assessment of every feature in the hosted application.
For consumers, the terms allow the provider to use content to provide, maintain and improve services. They say users can opt out of model training through account settings. They also state that content associated with user feedback or flagged for safety review may be used for training even after opting out. These exceptions are narrower than a blanket statement that opting out stops every use of content.
The privacy policy describes a different default for API and business services: customer content is not used to train foundation models by default, except for participation in a free, trial or promotional program when the customer has not opted out. The policy also describes use for safety, security and legal compliance. The privacy policy says personal data contained in content processed on an organization's behalf is handled under the organization's agreement and data-processing addendum; that organization's own privacy policy governs its use of the data.
The consumer terms state that users retain rights they have in inputs and receive any provider rights in outputs, subject to the terms. The same terms warn that outputs may not be unique. Users also represent that they have the necessary permissions for submitted inputs, so the service's assignment of any output rights is not itself proof of rights in material supplied to it.
The privacy policy describes collection of account identifiers, input/output content, feedback, device metadata and usage analytics. It describes purpose-based data retention and says some data may be retained after account deletion for legal or dispute reasons. It identifies service providers and other recipients as possible data recipients, and describes processing in the United States, Singapore and other operating countries.
