New independent test 108 responses examined

We tested 9 free AIs. The winner changed when we looked beyond the score.

Grok and Claude led the recorded capability scores. ChatGPT and Claude both passed every instruction-compliance check. Copilot’s answers were recoverable—but severe display problems affected all 12 tests.

Read the findings Help me choose

9 products · 12 identical tasks each · Evidence preserved

One score misses the storyWe measured capability, instruction-following and real free-tier usability separately.

The Free Tier Reality Check

Three measures. Three different stories.

A high-quality answer is only part of the experience. We kept capability, prompt compliance and practical access separate—so a polished average could not hide a failed instruction or a broken interface.

01 · Capability4.98/5

Grok recorded the highest mean

Claude followed at 4.96. The 0.02-point gap is too small to claim a definitive winner.

02 · Compliance12/12 each

ChatGPT and Claude passed every prompt check

Both products received 12 Pass classifications in the separate instruction audit.

03 · Usability12/12 affected

Copilot’s display failed repeatedly

Answers were often blank, offscreen or only partly visible, despite usually being recoverable through Copy.

The honest conclusion: this test does not produce one universal “best free AI.” The right choice changes with the thing you value most.

See all 9 products and the limitations

What would you like help with?

Choose a task, not a brand.

These early findings come from preserved, independently evaluated reviews.

I want to…

Edit a book

Recorded best fitClaude

Claude recorded 5/5 in both editing tests; Copilot also reached 5/5 for developmental critique.

Explore this result →

I want to…

Write a report

Recorded best fitClaude

Recorded 5/5; Microsoft Copilot followed at 4.75/5.

Explore this result →

I want to…

Create images

Recorded best fitGemini

Strongest recorded visual result across the tested tasks.

Explore this result →

I want to…

Try AI for the first time

Recorded best fitChatGPT, Claude or Copilot

All three recorded 5/5 for approachable first-time guidance.

Explore this result →

Early evidence, clearly labelled.Results reflect the tested models, plans and dates. AI products change, so we keep testing and updating.

Trust without the technical overload

Simple answers on the surface. Rigorous checks underneath.

We built AROS because choosing an AI should not depend on advertising, popularity or somebody’s personal favourite.

Read about our testing and governance →

Same tasks

Each AI receives the same controlled prompts.

Saved evidence

Prompts and complete responses are preserved.

Independent check

Results are assessed separately before publication.

Why we’re here

AI is moving quickly. Choosing it shouldn’t feel frightening.

AI Reality Check was created to help everyday people and businesses understand what different AI systems genuinely do well. We test practical tasks, keep the evidence and explain the outcome in plain English.

We tested 9 free AIs. The winner changed when we looked beyond the score.

Three measures. Three different stories.

Grok recorded the highest mean

ChatGPT and Claude passed every prompt check

Copilot’s display failed repeatedly

Choose a task, not a brand.

Edit a book

Write a report

Create images

Try AI for the first time

Simple answers on the surface. Rigorous checks underneath.

Same tasks

Saved evidence

Independent check

AI is moving quickly. Choosing it shouldn’t feel frightening.

Useful AI guidance, without the overwhelm.