Tools in ChatGPT, Claude, Gemini, what works, what doesn't
Web search, code execution, file uploads. What the big three can do and where the traps are.
What tools are
Tools extend the model with capabilities it doesn't have on its own: live web search, code execution, file ingestion, image generation.
What the three can do (state 2026)
ChatGPT Plus/Pro: Web search, Code Interpreter, file upload (PDF, Excel, Word), image generation, GPTs (custom apps), ChatGPT Agent (browser control).
Claude Pro/Max: Web search, Artifacts (live-rendered code/HTML), file upload, MCP servers (external), Projects (persistent context), Claude Code (terminal integration).
Gemini Advanced: Native Google search, Canvas (live-render), file upload incl. video, Gems (custom), Deep Research.
Where they differ
Code execution: ChatGPT's Code Interpreter is the most mature, Python sandbox with libraries. Claude's Artifacts are UI-strong. Gemini Canvas sits in between.
Web search: Gemini has the best index (Google), Claude is faster at summaries, ChatGPT has the best source integration.
File upload: All three handle PDFs. Claude handles very long documents best thanks to 1M context, Gemini handles long videos best.
External tools: Only Claude with MCP has an open standard. ChatGPT GPTs are proprietary. Gemini has nothing comparable.
The most common mistake
Believing the model searches automatically. With all three you have to explicitly say "search the web" or the model ignores the web capability and answers from training data (which is 6 to 12 months old).
The second most common mistake
Not checking the results. Web search returns sources, at least click through them as a spot check. Especially for news, numbers, quotes.
Practical tip
For long-term productivity: Claude with MCP. The open standard means you can attach your own tools and memory servers. That's the bridge to Level 4.