Step by Step
O
OpenAI — GPT series
OpenAI's GPT series includes GPT-4o for general reasoning, and the o1/o3 models specifically designed for extended reasoning on hard math and science problems.
Example: choosing an o-series model specifically for a challenging multi-step math proof, rather than GPT-4o, given its focus on extended reasoning tasks.
A
Anthropic — Claude series
Anthropic's Claude series, including Claude 3.5 Sonnet, is particularly strong for nuanced writing, coding, and analysis, offering a 200K token context window.
Example: choosing Claude 3.5 Sonnet specifically for a nuanced long-form writing or complex coding task, given its documented strengths in these areas.
G
Google — Gemini series
Google's Gemini series, including Gemini 1.5 Pro, is multimodal and offers a massive 1M token context window, large enough to include an entire codebase.
Example: choosing Gemini 1.5 Pro specifically for a task requiring analysis of an entire large codebase at once, given its 1M token context window.
M
Meta — Llama series, open-source
Meta's Llama series is open-source, free to run locally and fine-tune, with Llama 3.1 405B now competitive with GPT-4 on many benchmarks — pricing across the landscape ranges from expensive frontier APIs to fully free open-source options.
Example: choosing Llama 3.1 405B specifically because it can be run locally and fine-tuned freely, avoiding the ongoing API costs of a frontier proprietary model, while still achieving competitive benchmark performance.
Applied Walkthrough
1
A team is choosing among the four major LLM providers for different specific needs across their organization.
2
For a challenging multi-step math and science reasoning task, they choose an OpenAI o-series model, given its specific focus on extended reasoning.
3
For a task requiring analysis of an entire large codebase at once, they choose Google's Gemini 1.5 Pro, given its massive 1M token context window.
4
For a cost-sensitive internal tool where they want full control and no ongoing API costs, they choose Meta's open-source Llama 3.1 405B, which is now competitive with GPT-4 on many benchmarks while being free to run locally and fine-tune.
Exam Application
Exams test whether you can match each of the four major providers (OpenAI, Anthropic, Google, Meta) to their flagship models and documented strengths, and whether you know the specific fact that Llama 3.1 405B (open-source) is now competitive with GPT-4 on many benchmarks.
⚠ Common Trap
The most common trap is assuming open-source models are automatically inferior to proprietary frontier models. Llama 3.1 405B specifically is documented as competitive with GPT-4 on many benchmarks, showing that the open-source vs. proprietary distinction doesn't automatically determine which model performs better.
✓ Quick Self-Check
1. What are OpenAI's o1/o3 models specifically designed for?
Extended reasoning for hard math and science problems.
Tap to reveal / hide
2. What context window size does Claude 3.5 Sonnet offer?
200K tokens.
Tap to reveal / hide
3. What context window size does Gemini 1.5 Pro offer, and what does this enable?
1M tokens, enough to include an entire codebase.
Tap to reveal / hide
4. What is notable about Meta's Llama 3.1 405B?
It's open-source, free to run locally and fine-tune, and is now competitive with GPT-4 on many benchmarks.
Tap to reveal / hide
5. What is the general pricing range across this landscape of models?
From expensive frontier proprietary APIs to fully free open-source options.
Tap to reveal / hide