Simon Willison updated his Claude Token Counter to show token counts across different Claude models side by side. He published the details on his blog, and the update is small but practically useful for anyone calling Claude via API.
Tokens are the basic units Claude uses to process text - roughly three-quarters of a word each, so a 1,000-word document runs around 1,300 tokens. The count matters for two reasons: Anthropic charges per token, and every model has a maximum context window (the ceiling on how much text it can handle in a single request). A 200k-token context window holds roughly 150,000 words.
The complication is that different Claude models don't always tokenize the same text identically. The differences per prompt are small, but they add up across thousands of API calls, and they affect whether a long document fits inside a smaller model's context limit. Before this update, checking each model required separate requests.
The comparison view collapses that into one step. Paste your prompt, see the token count for each model simultaneously, and know your cost and context headroom before committing to a model choice.
For developers building directly on Claude or running prompt-heavy workflows in Claude Code, it's a faster way to catch unexpectedly long prompts before they inflate your API bill or silently truncate your input.