How many tokens is 1% of your Claude Code limit?

Description and use

Claude Code users lack a way to translate usage percentages into token counts, hindering budget management. The author measured that 1 % of the 5‑hour limit equals ~166k new tokens or ~11.2M cached tokens, with new tokens making up 2.5 % of all tokens but 63 % of the limit. This insight helps optimize subagent use, session length, and context size.

Problem

Users cannot directly translate Claude Code usage percentages into token counts, making it hard to gauge how much of their token budget is consumed by different token types.

Outcome

The study found that 1% of the 5‑hour limit equals roughly 166k new tokens or 11.2M cached tokens, with new tokens accounting for 2.5% of all tokens but 63% of the limit. Subagents, long sessions, and large starting contexts significantly impact token consumption.

Workflow

• 1. Record status‑line percent readings from Claude Code over time.
• 2. Extract token counts (input, cache writes, cache reads, output) from transcripts for each model call.
• 3. For each interval where the 5‑hour percent rose, compute the rise and split the tokens into new (input, cache writes, output) and cached (cache reads).
• 4. Perform a least‑squares fit (no intercept) of rise in % ≈ a × new tokens + b × cached tokens.
• 5. Derive % of the 5‑hour limit per million new and cached tokens and compute tokens per 1%.
• 6. Use the resulting ranges (≈165–200k new tokens, 10.7–11.2M cached tokens per 1%) to assess token usage.
• 7. Apply insights to optimize subagent usage, session length, and context size.
• 8. Optionally use the limitcast CLI to automate data collection and fitting.
• 9. Share findings to compare ratios across plans.

Evidence

1% of the 5-hour limit was about 166,000 new tokens, or about 11 million cached tokens.

Source

New tokens (input, cache writes, output) 2.5% of everything Claude Code sent, and they used about 63% of my limit.

Source

Subagents sent about a third of my tokens (33%), but nearly half of my new tokens (48%), and used an estimated 42% of my limit.

Source