CodeCondense
13 real tests · 2 scenarios · every token measured

Slim cuts Claude's overhead by up to 83%

Two test suites. Same model. Every tool call, token, and second measured.

Single session · 10 build tests

−85%

Fewer calls

−72%

Fewer tokens

$1.44

Cost saved

15m 46s

Time saved

Across 10 apps — Node.js + Python, build from scratch

8 h/day · 20 working days · one large repo

$631

saved per month

$125 with Slim vs $756 vanilla

19h

Hours returned

of dev time/month

20M

Tokens saved

tokens/month

Methodology: Every task ran twice — once vanilla Claude, once with Slim. Tool calls, tokens, and wall-clock time recorded from Claude Code telemetry. Pricing: Claude Sonnet 4 at $3/M input · $15/M output. Claude Sonnet 4claude-sonnet-4-6Slim v0.8.36June 2026

Project complexity

Next.js 15 legal AI SaaS (private repo)

7,256

Total files

1,326 .tsx + 1,699 .ts

783

Components

avg 290 lines each

192

Dependencies

prod packages

4,375 lines

Largest file

171 MB total

95+ API routes across feature domains
Files spread 5 levels deep — find without Search = repeated Bash calls
4,375-line files: vanilla must read thousands of lines to locate injection points

−72%

Fewer calls

3 modification tasks

−82%

Fewer tokens

3 modification tasks

$7.18

Cost saved

3 modification tasks

13m 12s

Time saved

3 modification tasks

All 3 tasks — side by side

Tool calls

SimpleAdd scroll-to-top button to landing…
−63%
Vanilla
8
Slim
3
MediumAdd "Export to CSV" button…
−71%
Vanilla
7
Slim
2
ComplexAdd "Recent Activity" feed to…
−79%
Vanilla
14
Slim
3

Tokens used

SimpleAdd scroll-to-top button to landing…
−71%
Vanilla
38K
Slim
11K
MediumAdd "Export to CSV" button…
−83%
Vanilla
95K
Slim
16K
ComplexAdd "Recent Activity" feed to…
−85%
Vanilla
145K
Slim
22K

API cost (USD)

SimpleAdd scroll-to-top button to landing…
−82%
Vanilla
$1.20
Slim
$0.22
MediumAdd "Export to CSV" button…
−84%
Vanilla
$2.80
Slim
$0.44
ComplexAdd "Recent Activity" feed to…
−83%
Vanilla
$4.60
Slim
$0.76

Wall-clock

SimpleAdd scroll-to-top button to landing…
−66%
Vanilla
3m 5s
Slim
1m 2s
MediumAdd "Export to CSV" button…
−83%
Vanilla
4m 45s
Slim
48s
ComplexAdd "Recent Activity" feed to…
−83%
Vanilla
8m 40s
Slim
1m 28s
Vanilla ClaudeWith SlimT1: SimpleT2: MediumT3: Complex

Key finding: token penalty grows with file size

Task 2 involved a single 2,478-line file. Vanilla must Read up to 2,000 lines to locate the injection point at line 1,049 — loading ~8,000 tokens of irrelevant context. Slim's Search returned the 20 relevant lines in one call, cutting input tokens by 74%.

In large codebases, this pattern repeats on every file you touch — savings compound across sessions

Combined totals — 3 modification tasks

MetricVanillaSlim
Total tool calls29
8−72%
Total tokens278K
49K−82%
Total API cost$8.60
$1.42−83%
Total time16m 30s
3m 18s−80%

$7.18 saved across 3 real-world modifications

72% fewer calls · 82% fewer tokens · 13m 12s returned

Monthly projection

What does this cost at scale?

8 hours a day in a large codebase, 20 working days. Here's where the money goes.

Cumulative API cost — 8 h/day × 20 days

15 tasks/day · mixed complexity

Cumulative API cost over 20 working days

$756$125$200$400$600$756Day 1Day 5Day 10Day 15Day 20
Vanilla ClaudeWith Slim

Monthly totals

API cost−83%
Vanilla
$756
Slim
$125
Tool calls−70%
Vanilla
2,640
Slim
780
Tokens−82%
Vanilla
24.7M
Slim
4.6M
Time waiting−79%
Vanilla
24.3h
Slim
5.1h

$631

Monthly cost saved

$125 with Slim vs $756 vanilla

1,860

Fewer tool calls

780 vs 2,640 / month

20M

Tokens saved

4.6M vs 25M

19h

Time returned

5.1h vs 24.3h waiting

Assumptions: 15 Claude Code tasks per 8-hour day in this 7,256-file codebase (40% simple / 40% medium / 20% complex). Vanilla averages $2.52/task — dominated by repeated full-file reads and cumulative context re-sends. Slim averages $0.42/task via targeted Search + batch Edit. Pricing: Claude Sonnet 4 at $3/M input · $15/M output.