Get Started
← Own your AI · Mission 2 of 31

The tiny AI model that got basic maths wrong, and the one to use instead

Ask a tiny free model when a 14:35 train arrives after 2 hours 50 and it confidently says 15:25. Find the model that gets it right on your laptop, plus a one-command test that checks any model before you trust it.

17 minBeginnerTools: FreeStep 1 free
Tested 7 Oct 2026 · Ollama 0.40, Gemma 4 e2b, Qwen 3.5 0.8b and 4b, 16GB Intel laptop, no graphics card
A clear rule of thumb for which model size suits your machine
Real resultsspeed, size and right or wrong answers for three popular models
A one-command, 4-question trust test that marks any model for you
Before you start
  • The first lesson in this track done: Ollama installed. Gemma 4 needs Ollama 0.20.7 or newer: check with ollama --version.
  • Up to 5GB of free disk space per model you try
  • Python, or uv, for the one-command trust test (step 4). Not needed for the rest
01/04Understand what the size means0 done
STEP 01 of 04~3 min

Understand what the size means

The number in a model's name tells you roughly how much it knows and how hard your laptop will work.

  1. Model names end in a size like 0.8b, 4b or e2b. The b means billions of parameters, the model's "brain cells". Bigger usually means smarter, slower and a larger download.

    • Under 1b: tiny and quick, fine for simple rewording, unreliable for facts and maths.
    • 2-4b: the sweet spot for most laptops.
    • 7b and up: noticeably smarter, but you'll want 16GB+ of RAM, ideally a graphics card.

  2. Which model for your RAM (a starting point from the results in step 2):

    • 8GB: qwen3.5:4b (3.3GB download). Slower, but it follows instructions well. Close other programs while it runs.
    • 16GB: gemma4:e2b (4.6GB). The best all-rounder in this test.
    • 32GB or a graphics card: start with gemma4:e2b, then try the bigger sizes in the same families (gemma4:e4b, qwen3.5:9b) with the trust test in step 4 before you switch.
    TipGemma 4's e2b means effective 2b: the download includes extra parts that don't all need to be worked through for every word, so it runs more like a 2b model. That's why it runs faster than the 4b model, despite being a bigger download.
✓
CheckpointYou can say what 4b means and which size range suits your laptop.
Stuck? Ask the community
Next mission · Own your AIYour own private ChatGPT, in your browser
→