ModelCompareCompare models

ModelCompare blog

We test AI models so you don't have to guess

We send the same prompts to many models and publish every answer and what it cost, then write up where they actually differ.

6 stories · Latest October 18, 2026

GPT vs Claude

Which AI model glazes you the most?

We texted GPT and Claude models in slang, emoji, formal English and two-word questions, then asked them to rate our work. GPT copies your emoji. Claude brings its own.

Read the story

+1.38

points GPT-6.1 Sol adds to a 1 to 10 rating when you say you wrote it

More stories

  1. Image models

    4 of 27

    scored image models worth their price; every other one has something cheaper that scores higher

    You don’t need the top AI image model

    We mapped 35 image models by arena quality and price, then ran the same prompts, edits and borderline content tests through the leaders and the best open-weight models.

  2. The race

    0

    days an open-source model has held the top score since ChatGPT launched

    Open-source AI is only four months behind

    Since ChatGPT launched, a closed AI model has held the top score every day. We charted 694 models to see how far behind the open ones are, and where they lead.

  3. The test

    5

    tests that still split four cheap models, out of 884 conversations

    Cheap AI got good. Here's what still trips it up

    We ran 884 conversations through GPT-6 Luna, Claude Haiku 5.5, GLM 5.3 Flash and Gemini 3.8 Flash. They passed almost everything. Five tests still split them.

  4. Guide

    30

    models sorted by family, size and price, updated with the catalog

    AI models explained: names, prices and how to choose

    A plain-English guide to AI model families like Claude, GPT and Gemini, how per-token pricing turns into a real bill, and how to choose.

  5. Same prompts, real costs

    21.6×

    Muse Spark Open's bill compared with GPT-6 Luna's for the same 241 prompts

    Gemini Flash vs Haiku vs Luna vs Muse: same prompts, real costs

    We sent the same 241 prompts to four cheap models: drawings, games, landing pages, puzzles and 40 political statements. Every answer, and what each one cost.

Run the test on your own prompt

Send one prompt to up to 6 models and see every answer, its speed and its cost.

Compare models