LLM Guides Lab

LLM Guides Lab: hands-on LLM tests, published in full

We test ChatGPT, Claude and Gemini on real everyday tasks, score every answer blind, and publish all of it: the prompts, every answer, every score. Then we test again as the models change.

Browse the tests How we test

1test published

3models compared

24answers published

1test dates since Oct 2026

Latest: Writing Test publishedOctober 8, 2026

Real tasks

Prompts people actually type at work, not synthetic benchmarks.

Scored blind

We rate every answer before we know which model wrote it.

Everything public

Every answer and score is on GitHub, unedited.

Retested over time

Same prompts each time, so you can see who’s improving.

Who leads right now

Average score across our quality tests, and the leader in each one.

Full leaderboard →

Provisional

Claude

4.8 / 5

1 quality test

Provisional

ChatGPT

4.6 / 5

1 quality test

Provisional

Gemini

4.1 / 5

1 quality test

Latest score per model in each LLM Guides Lab test
TestClaudeChatGPTGeminiLast tested
Writing 4.8 4.6 4.1 Oct 2026
Scores out of 5. Measured tests (like speed) are shown in their own unit and aren’t included in the average. "=" marks a tie (lead under 0.2). Models are provisional until they have results in three quality tests.How we score →

All tests

Each test page shows every prompt, every answer and how the results changed over time.

Test 01

LLM Writing Quality Test

Eight writing tasks, from a blog intro to landing-page copy. ChatGPT, Claude and Gemini answered on the same day, and every answer was scored blind.

★ Claude leads · 4.8

Claude4.8
ChatGPT4.6
Gemini4.1

Tested 1 timeUpdated Oct 2026

Don’t miss the next head-to-head.

We put ChatGPT, Claude and Gemini through real tasks and show everything. Get the next result in your inbox.

No spam. Unsubscribe anytime. Privacy policy

Latest updates

  • October 8, 2026LLM Writing Quality Test Writing test, October 2026: Claude leads with 4.8, ahead of ChatGPT 4.6 and Gemini 4.1

Open data

Every prompt, answer and score from every test date is published on GitHub as plain text and CSV. Check our numbers, or use the data in your own work.

Answers published24

Scores recorded120

Test dates1

Data licensed CC BY 4.0: free to reuse with credit to LLM Guides.

Browse the data

How every test works

The same process for every test, every time.

Full methodology →

  1. Same prompts

    Fixed prompts, fresh chats on paid plans, all three models on the same day. Web search on; memory, chat history, custom instructions and connected apps off.

  2. Answers locked first

    Every answer is published to GitHub before any scoring starts.

  3. Blind scoring

    Rated 1–5 on five criteria without knowing which model wrote it.

  4. Retest and compare

    We rerun the same prompts as models change and show what moved.