# Can You Cut 80% of Claude's System Prompt?

> Published 2026-07-25 · https://www.promptzone.com/maeve_rahimi/can-you-cut-80-of-claudes-system-prompt-1c2c

Anthropic's Claude Code system prompt was trimmed by more than 80% for two internal coding projects, Opus 5 and Fable 5, according to [a recent Hacker News thread](https://twitter.com/trq212/status/2080710971228918066).

The change kept core instruction quality while dropping token count sharply.

> **Model:** Claude Code | **Reduction:** 80%+ | **Projects:** Opus 5, Fable 5
> **Source:** Hacker News discussion (16 points, 2 comments)

## What the Trimmed Prompt Achieves

The original Claude Code prompt contained extensive rules for formatting, safety, and workflow. The reduced version kept only the minimal constraints needed for code generation and review.

Early reports indicate the shorter prompt maintained output consistency on both Opus 5 and Fable 5 tasks.

## How the Reduction Was Measured

The team reported removing over 80% of tokens from the system prompt. The remaining instructions focused on output format and basic safety boundaries.

No public benchmark numbers were shared beyond the length claim.

## How to Test a Similar Reduction

Start with your current Claude system prompt. Identify sections that repeat across responses or restate model defaults.

Remove one category at a time—formatting rules, example chains, or safety reminders—and run the same coding task set before and after.

Track token usage and output quality on a fixed test suite of 10-20 prompts.

## Tradeoffs Observed So Far

- Shorter prompts lower cost per call and reduce latency.
- Risk of drift increases if critical constraints are dropped.
- Projects with strict style guides may need at least one retained formatting rule.

## Alternatives to Full Prompt Trimming

Teams often compare three approaches:

| Method              | Token Savings | Risk Level | Best For                  |
|---------------------|---------------|------------|---------------------------|
| Full 80% cut        | 80%+          | Medium     | Internal tools            |
| Selective pruning   | 30-50%        | Low        | Production APIs           |
| Few-shot examples   | 10-20%        | Low        | Strict output formats     |

Claude's own prompt caching and Anthropic's newer context management features offer partial savings without manual editing.

## Who Benefits Most

Developers running high-volume code generation inside Claude Code see the clearest cost drop. Teams with heavy compliance requirements should keep more of the original prompt.

Researchers testing prompt efficiency on Opus-class models can use the 80% figure as a starting benchmark.

## Practical Next Steps

Export your current system prompt. Mark every sentence that has triggered at least once in the last 100 calls. Delete the rest and retest.

Measure both token count and pass rate on your standard coding tasks before adopting the shorter version.

> **Bottom line:** An 80% system prompt reduction is achievable for focused coding projects when only essential constraints remain.

The result points to continued pressure on teams to measure every token against measurable output quality.