---
title: "Study 6: four people, twelve AI coding agents, 29% fewer tokens · Cork"
description: "Four people running twelve Claude Code agents at once on one repo, four rounds with Cork and four without, same tasks and model: 13.78M vs 19.31M tokens in total, 29% fewer (17 to 39% per round), 47 of 48 tasks correct vs 47."
canonical: "https://corkcli.com/benchmarks/study-6-four-people-twelve-agents"
last_updated: "2026-11-14"
---

# Four people, twelve agents: 29% fewer tokens with Cork

Twelve Claude Code agents launched at once on one repository, each with its own task, all pushing to the same origin. We ran it four times with Cork and four times without: same tasks, same prompts, same model. Over the four rounds the agents with Cork used 13.78M tokens and the agents without it used 19.31M. Every round saved between 17% and 39%.

29%
:   fewer tokens, four rounds

18%
:   fewer turns (609 vs 746)

18%
:   lower cost ($8.50 vs $10.40)

47/48
:   tasks correct, vs 47/48 without

## Every round

Tokens for all twelve agents in each round, bars to scale.

Columns: with Cork, without

Round 1 · 32% fewer

3.12M

4.57M

Round 2 · 24% fewer

3.52M

4.65M

Round 3 · 39% fewer

3.44M

5.61M

Round 4 · 17% fewer

3.71M

4.48M

## The numbers

Tokens are input, cache writes, cache reads and output, as the provider reported them for each agent. Cost is the model's list price for that usage. Correct means the task's hidden acceptance tests pass on the final shared origin.

| Round | With tokens | Cost | Correct | Without tokens | Cost | Correct |
| --- | --- | --- | --- | --- | --- | --- |
| Round 1 | 3,118,659 | $1.94 | 12/12 | 4,568,882 | $2.46 | 12/12 |
| Round 2 | 3,518,965 | $2.17 | 12/12 | 4,651,576 | $2.57 | 12/12 |
| Round 3 | 3,435,698 | $2.14 | 12/12 | 5,608,755 | $2.92 | 11/12 |
| Round 4 | 3,711,600 | $2.25 | 11/12 | 4,483,969 | $2.45 | 12/12 |
| All four | 13,784,922 | $8.50 | 47/48 | 19,313,182 | $10.40 | 47/48 |

## How it was run

- **The work.** Study 3's hardest tier: twelve tasks in a 654-file Python repository, several of them vague tickets, with decoys and one task that depends on another agent's work. Each agent got its own clone and pushed to one shared origin.
- **With Cork.** Cork's paid plan, set up the way a team sets it up: each person signed in on their own machine and ran the one setup command it returned, which installed Cork's hooks for that machine; each agent also had Cork's MCP server under its person's sign-in. It ran on a staging copy of Cork's production servers, on the same build.
- **Without.** The same agents, prompts and flags with no Cork.
- **Both.** Claude Sonnet 5, all twelve agents launched at once, three per person. Any agent whose work had not reached the origin when it stopped got one identical wake-up message, in both groups.

## What this study does not show

- The four people were test accounts driven by one harness, and all twelve agents started at the same moment rather than over a working day.
- Agents finished in about a minute each, so these rounds say little about wall-clock time.
- One task failed once with Cork (round 4) and one failed once without it (round 3).
- In dollars the gap is smaller than in tokens (18% vs 29%), because much of each session is billed at the cache-read rate.

[All studies](https://corkcli.com/benchmarks)[Install Cork](https://corkcli.com/install)

This page in HTML: https://corkcli.com/benchmarks/study-6-four-people-twelve-agents