Submit tool

ChatGPT alternatives

How we compare ChatGPT alternatives.

By · tested and updated September 2026

This site is built around search intent and practical workflows, not a single artificial score. A good ChatGPT alternative for research may be a weak choice for presentations; a strong writing model may be the wrong tool for source-backed answers.

Browse all guides Back to homepage
Short verdictMethod
Best forWriting, research with sources, coding, file upload, document summary, presentations, images and business workflows
Check before payingLimits, data and workflow

Quick answer

We compare tools by what users actually need: writing, research, coding, files, images, privacy, business fit, pricing transparency and whether the answer can be verified.

Do not take any comparison site's word for it, this one included. Read the method, check a source link, and see whether the two agree — that is a two-minute test of whether a site is worth your time.

The full test protocol is published separately: the 20 prompts, the scoring rubric and the recording rules are fixed in public before any model is run, so the method can be checked independently of the results.

Decision map

What to check before choosing.

What we test

What we test

Writing, research with sources, coding, file upload, document summary, presentations, images and business workflows.

What we avoid

What we avoid

Fake exactness, unverified prices, pretending AI answers are always correct, and claiming detector or quality guarantees.

How to read the site

How to read the site

Start with your task, then compare two or three tools using your own prompts before paying.

Rule by rule

What we hold ourselves to, and what we do not claim.

AreaUseful forWatch out for
Real promptsWe prefer workflow tests over generic opinions.Outputs change, so tests are examples, not permanent truth.
Official sourcesPricing, privacy, file limits and plan names should be checked on official pages.Third-party summaries can be outdated.
Use-case scoringA tool can win one use case and lose another.There is no universal winner for every person.
DisclosureSome pages mention MultipleChat because it is relevant to multi-model workflows.Recommendations should still be useful even if the reader never opens it.

How this site is actually built

Four rules we hold ourselves to.

A methodology page that only describes intentions is decoration. These are testable claims about how the pages here are made, and you can check each one against any page on the site.

Every figure carries a source and a date

Prices, context windows and limits on this site link to the vendor's own documentation and state when we read it — currently 2026-09-10. If a figure has no source link, it should not be here; tell us and we will fix it.

We publish the gaps

Where a vendor does not publish a number, we say so rather than copying one from a comparison blog. Google's per-model token limits and several vendors' current plan limits are marked as unverified for exactly this reason.

The test method is public before the results

Our twenty-prompt protocol, rubric and recording rules are published in full at the test protocol page, with an empty results table and a date. Publishing the instrument first is what makes the eventual numbers checkable.

No paid placement

Nobody pays to appear on this site or to be ranked higher on it. Where several tools do the same job, we name all of them rather than only the one that suits us.

What we do not claim: that we have run controlled benchmarks, that our recommendations are measured results, or that any ranking here is permanent. Where a recommendation is a preference formed from our own use, the page says so in those words.

What would change our mind

The conditions we would revise on.

A published recommendation should come with the conditions under which it fails. Ours: if our own test protocol produces results that contradict a recommendation on this site, the recommendation changes and the change is dated rather than quietly edited.

If a vendor publishes specifications that contradict a figure here, the source wins and we correct the page. If you find one, the fastest route is the submit form — we would rather be corrected than consistent.

Sources

Where these claims come from.

Every figure on this page was read from the official documentation below on 2026-09-10. Prices, limits and model names change without notice — the source is authoritative, this page is not.

OpenAI model documentation: current model names, context windows and output limits. Anthropic model documentation: Claude model IDs, context windows and max output tokens. Google Gemini model documentation: the current Gemini model family. Anthropic pricing: current Claude plan prices and usage limits. OpenAI pricing: what each ChatGPT tier includes.

FAQ

Questions about our method.

How do you decide which tool is best?

We do not, in the general case — there is no best, only best for a task and a constraint. Each page states what it is optimising for, and where a view is a preference rather than a measurement, it is labelled as one.

Where do your numbers come from?

The vendors' own documentation and pricing pages, read on a stated date — currently 2026-09-10 — and linked from the sources section on every page. We do not source figures from other comparison sites.

Do you accept payment for placement?

No. No tool on this site has paid for placement or for a higher ranking, and no ranking here is for sale.

Why are some numbers missing?

Because we could not confirm them at source. Google does not publish per-model token limits on its overview page and several vendors do not publish current plan limits. Printing an unverified figure would make the whole table less trustworthy, not more complete.

How often is this updated?

Figures are re-checked and re-dated rather than silently refreshed, and the test protocol runs quarterly. If a page's date looks stale to you, that is the correct signal to follow the source link instead of trusting the page.