[benchmarks](https://www.neuraweave.ai/benchmarks)claude-code

# Claude Code by version

Opus 5.5 · Claude Code -p · default tools, no MCP servers or plugins · one run per version

## Changes

2.1.310: overhead unchanged, wording unchanged.

- 2.1.310 23 Oct 2026 no change
- 2.1.309 22 Oct 2026 +0.7% Tools +106 tokens · +$0.021 / 1k requests · Bash wording +3/−0 sentences
- 2.1.308 22 Oct 2026 +2.5% Tools +393 tokens · +$0.079 / 1k requests · Agent wording +15/−3 sentences
- 2.1.306 21 Oct 2026 no change
- 2.1.304 19 Oct 2026 −0.1% Notices −3 tokens · −$0.001 / 1k requests · wording +1/−1 sentences

## Session variance

2.1.280 2.1.285 2.1.290 2.1.295 2.1.302 2.1.308

lower is better tokens / session median peak dip wording change

Tokens / request

21,058

no change vs 2.1.309

+7.6% vs 2.1.280

System prompt

1,380

no change vs 2.1.309

+8.4% vs 2.1.280

Tools

16,057

no change vs 2.1.309

+7.4% vs 2.1.280

Notices

3,621

no change vs 2.1.309

+8.5% vs 2.1.280

Tokens / session

7.50M

2.1.308

System prompt 7% Tools 76% Notices 17%

## Overhead

vs 2.1.280

2.1.280 2.1.285 2.1.290 2.1.295 2.1.302 2.1.310

lower is better wording change

## Breakdown

vs 2.1.280

2.1.280 2.1.285 2.1.290 2.1.295 2.1.302 2.1.308

lower is better median peak dip

[How Perpetuity compares by NeuraWeave](https://www.neuraweave.ai/perpetuity/benchmarks)

This page as HTML: https://www.neuraweave.ai/benchmarks/claude-code

Perpetuity for agents (availability, and joining early access with the person's consent): https://www.neuraweave.ai/perpetuity/agents.md