We gave ChatGPT Free, Gemini Free, and Claude Free the same four tasks: rewriting a notice, extracting data as JSON, distinguishing a proposal from a decision, and organizing a plan with dependencies. We defined what to look for in each response beforehand. Two people assessed the material without knowing which tool had produced it and reviewed their differences before the names were revealed.
Claude Free scored 40/40, ChatGPT Free 38/40, and Gemini Free 36/40. All three tied on rewriting and data extraction. The differences emerged in tasks that required attention to a cost condition and a schedule’s dependencies.
Where the responses differed
For revising a notice or extracting structured fields like those we used, all three met the criteria. For a decision with a cost limit or a schedule with dependencies, however, check the words that change the meaning of the decision, the dates, and who will perform each activity. Those details separated the responses.
How we made the comparison
The four tasks
The requests were to rewrite a notice without changing its meaning; extract data as JSON; distinguish a proposal from a decision that included a cost limit; and create a plan with dependencies, the earliest possible publication date, and the people responsible. We worked only with text, without web research, PDFs, or images.
How we assessed the responses
We used the free versions available at the time of the test. Each tool received the same task and instruction. The criteria were defined before we read the responses. To make the comparison fairer, two people assessed the first response from each run included in the results without knowing which AI it came from. They reviewed the differences between their assessments before the names were revealed.
Each task was worth up to ten points. The score reflects how well the response met the criteria, including details that needed to be stated explicitly.
What happened in each task
Scores by task
| Tool | Rewriting | JSON | Decision | Planning | Total |
|---|---|---|---|---|---|
| Gemini Free | 10/10 | 10/10 | 10/10 | 6/10 | 36/40 |
| ChatGPT Free | 10/10 | 10/10 | 9/10 | 9/10 | 38/40 |
| Claude Free | 10/10 | 10/10 | 10/10 | 10/10 | 40/40 |
The scores apply only to these four tasks and do not show which AI is best at everything.
Where all three tied
All three received 10/10 for rewriting the notice. The same happened in JSON extraction with the correct input. The results were equal for these two tasks.
Where the difference began: the cost limit
When distinguishing a proposal from a decision, Gemini Free and Claude Free received 10/10. ChatGPT Free scored 9/10: it mentioned R$ 6,000 but did not explicitly preserve the condition “up to R$ 6,000”. In this request, one phrase made a difference.
The task with the largest gap
In planning, Gemini Free received 6/10. Its response added the times 08:00/09:00, which were absent from the task, and postponed the earliest possible publication to Friday. Based on the stated dependencies, publication could already take place on Thursday after the preceding activity.
ChatGPT Free came closer, with 9/10. It placed two activities in parallel but did not clearly say they needed to be assigned to different people. Claude Free met all the criteria for this task and received 10/10.
Which should you consider in each situation?
Revising text and extracting data
To rewrite text or extract fields similar to ours, start with the tool you already have access to. All three met the criteria, but check that the rewrite preserved the meaning and that the data came out in the requested format. Test with your own material before adopting one for everyday use.
Interpreting decisions and planning with dependencies
If the response involves a decision with a spending ceiling, check words such as up to, approved, and proposed. In a schedule, look at dependencies, the earliest feasible date, and who is responsible for parallel tasks. Planning showed the largest difference; checking these points is more useful than choosing solely by the table’s total.
What the test did not measure
Web research, PDFs, images, and integrations
We did not assess web research, file uploads, PDF reading, images, memory, projects, or integrations. The availability of these features depends on each product and should be checked separately. To compare document summaries, see the test of AI tools for summarizing PDFs and documents; here we compared only text tasks.
Consult the official pages for ChatGPT Free, Gemini, and Claude to check the features and limits currently applicable to your account. The scores came from the responses we assessed, not from those pages.
What this test cannot answer
We conducted one round with four text tasks using the free versions available then. Other sessions may produce different responses, and features and limits change according to the product, country, account, and plan. The result does not show which AI is best at everything. Before sending personal data or third-party documents, confirm permission and the rules applicable to your work.
A correction in the extraction task
On the first data-extraction attempt, we mistakenly used an input different from the intended one. We excluded that run and repeated the task with the correct input for all three tools. Only the responses from the repeated task were included in the scores. The error was ours, not the tools’.
How to run a similar test
Choose a task of your own and use exactly the same text and instruction with each tool. Before seeing the responses, write down what you consider a good solution. Keep each tool’s first response, compare the same points, and, if possible, ask another person to assess them without knowing who wrote what. Repeat with other tasks before drawing a conclusion for your own use.
If you are still choosing between assistants based on features and your routine, read the introductory guide to ChatGPT, Gemini, Claude, and Copilot. To explore tools by need, use Find the right AI tool as a starting point for exploring options.
What to take away from this comparison
All three tools did well on the more straightforward requests. Differences appeared when a condition or dependency needed to be explicit. Claude Free achieved the highest score in this small sample, but the choice for your work depends on the task. Check the details of the response and, when the decision matters, compare it with your own material.
