OpenAI Dots vs Perplexity Computer: Research and Work

OpenAI dots vs Perplexity Computer is a useful comparison when a research question becomes an ongoing project. You may need more than a cited answer: a workbook, a briefing, a dashboard and later updates. Decide which output and sources matter before choosing the agent that will produce them.

This article uses public official pages checked on 30 September 2026. It offers a trial method rather than invented accuracy or speed results. An AI agent and an LLM are different layers; comparing products means checking the surrounding workflow as well as the model.

Search, Computer and dots are different entry points

Perplexity’s product overview separates its Answer Engine from Computer. It presents the former around cited answers and the latter around research, analysis, creation and coding in a multi-step workflow. If all you need is a source-backed answer, you may not need to launch a larger project.

OpenAI’s dots announcement describes agents that keep working across connected apps and ongoing goals. This makes dots a candidate when the responsibility extends beyond the first research output. It does not establish that dots research is more accurate than Perplexity’s, or the reverse.

The distinction to test is where the work should live. Is it a research project with an explicit deliverable? Is it a continuing responsibility that includes research, coordination and follow-up? Those are useful organising questions, but not exclusive capability categories. Both vendors describe broader workflows than a simple chatbot reply.

What Perplexity documents for project work

The official Computer at Work guide page describes research across web sources, uploaded files and connected apps, plus documents, spreadsheets, presentations, dashboards and recurring work. Its examples also distinguish local work using Personal Computer for Mac. The public page outlines the guide; this article does not pretend to have downloaded gated material.

That documented range supports evaluating Computer for a research-to-deliverable assignment. It does not tell you that every listed source, connector or local feature is available in every plan or country. Start by checking the actual account’s controls and supported connections, particularly where a job depends on proprietary data.

OpenAI’s dots tasks and memory documentation explains assigned responsibilities, delegated work and recurring tasks. Read it when the research output needs to become part of a continuing assignment. A schedule and a changing project are related, but a recurring run still needs clear sources, timing and a destination.

Pricing needs a workload, not just a monthly figure

Perplexity’s official pricing page shows individual Pro and Max monthly cards at $20 and $200, respectively, and distinguishes their Computer access and credits. It also advertises bonus credits. Check the current checkout, billing period and allowance; a promotional bonus should not be treated as a permanent monthly budget.

For dots, OpenAI’s access guidance describes the first dot as included in eligible Pro or Business Premium access, with an allowance for deeper work. It is not a separately priced unlimited research plan. Eligibility and gradual rollout need checking before cost comparisons are meaningful.

Keep separate records for the subscription, the task’s usage and any paid source used during the work. A credit is a billing unit, not a guaranteed number of completed reports. Do not infer equal output from equal spend across providers. One messy dataset can require more investigation than a folder of well-labelled exports.

If a trial uses public pages and copied sample files, review its usage before connecting expensive research sources or starting a recurring task. The right budget depends on your acceptable result, including the corrections needed to reach it. There is no measured saving in this article to substitute for that evidence.

Citations help you check, but do not finish the checking

A link beside a statement helps you find its evidence. It does not prove that the source supports the exact statement, that its date is relevant or that the agent has interpreted it correctly. Ask for the passage or calculation that changes the recommendation, not merely a long bibliography.

For supplier research, separate an official product specification from a review, a sales claim and an inference. For an internal report, keep filenames and definitions beside calculations. How to check an AI answer explains how to prioritise these checks when you cannot personally assess every detail.

Require the agent to leave a question unanswered when evidence is missing. A blank comparison field labelled “not verified” can be more useful than a fluent guess. This is especially important with regional availability, contract terms and prices that change by plan. The result should make uncertainty easier to see, rather than hide it in a confident summary.

A worked example: selecting a business software supplier

Imagine you are preparing a shortlist of software suppliers. You need official product pages, compatibility requirements and a draft evaluation workbook. The final purchase belongs to your organisation. This makes a suitable public-information trial for both products without uploading sensitive contracts.

Use a brief like this:

Compare the suppliers on the official pages I provide. Record source URLs and dates beside each verified capability. Keep prices in their original currency and billing period. Mark missing answers. Produce a draft workbook and a short recommendation based on our requirements. Do not contact vendors or create accounts.

For Computer, inspect the transition from research to the workbook and briefing. Do the same requirements appear in both? Can you trace a recommendation to a supported row? If a supplier makes an ambiguous claim, does the report preserve the ambiguity or translate it into certainty?

For dots, add the continuing responsibility: track the shortlist while you resolve unanswered questions, and revise the recommendation when you supply new evidence. How to use OpenAI dots explains a bounded first assignment. Do not give a general instruction to “manage procurement” when you only intend it to prepare research.

Introduce a revision that matters: a supplier’s connector supports reading but not writing, or its advertised price requires annual billing. Check whether the agent changes the recommendation and flags the affected output. A cosmetic update to the paragraph is insufficient if the underlying decision still uses the old assumption.

This is a proposed evaluation, not a report that we ran it. You can record the tools’ behaviour yourself: unsupported claims, missing requirements, useful questions, corrections and final output quality. Those observations are more informative than an unsourced winner label.

Connected apps change the scope

OpenAI’s computers and apps guide distinguishes cloud browser sessions, optional local access and plugins. An account connection supplies a source or tool; it does not establish what you want sent to another person. Check the actual account and destination before allowing a follow-up action.

Perplexity’s guide describes connected apps and local-file work, but a public capability list does not determine your organisation’s data permissions. If your trial later needs internal material, verify the plan, controls and approval process with the person responsible for that data. You can evaluate a workflow on public material before making this broader decision.

If most of the assignment is an office deliverable from your own files, dots vs Claude Cowork may be the more relevant comparison. If it becomes a repeatable operating process, building an AI workflow helps identify the inputs and decision points first.

Which should you evaluate first?

Evaluate Computer first when your main need is research turned into a specific deliverable and its available plan matches the project. Evaluate dots when the research is one part of an ongoing responsibility you want to steer in ChatGPT. Neither recommendation establishes a performance advantage; it gives your trial a purpose.

Coursium’s mobile learning approach is an option for learning how to brief AI and review its evidence. It does not replace a research subscription or run these projects for you. Start with a decision that you can check, preserve the sources, and choose the tool whose actual output helps you make that decision responsibly.