← All comparisons
The DIY alternative, not a tool

What Caveman gets you, vs. LemonCrow's runtime.

Caveman: a one-line terse-persona system prompt, no install. Tested head-to-head against LemonCrow's runtime on the same 20-prompt Telegraphic Q&A set.

The instruction itself
“Respond terse like smart caveman. All technical substance stay. Only fluff die.”
Free, zero install 47% avg output-token cut, 16pp stdev -- consistent Compresses replies only -- LemonCrow optimizes input tokens too, not just output
All 20 prompts -- in/out tokens, turns, cost -- median of 5 reps, same run as the flagship page (LemonCrow column highlighted)

"In" = summed prompt tokens (input + cache read + cache write). Split in BENCHMARKS.md.

Prompt Baseline LemonCrow Caveman
React re-render
object prop
20,811 in / 967 out
2 turns · $ 0.068
20,542 in / 473 out
1 turns · $ 0.039 ✓
22,792 in / 348 out
2 turns · $ 0.072
Express JWT expiry bug
benchmarks
70,873 in / 3,746 out
4 turns · $ 0.269
20,250 in / 2,908 out
1 turns · $ 0.099 ✓
22,821 in / 3,374 out
2 turns · $ 0.148
Postgres connection pool setup
benchmarks
20,803 in / 1,890 out
2 turns · $ 0.091
20,230 in / 1,138 out
1 turns · $ 0.055 ✓
22,785 in / 1,136 out
2 turns · $ 0.092
git rebase vs merge
benchmarks
20,804 in / 1,000 out
2 turns · $ 0.069
20,227 in / 622 out
1 turns · $ 0.043 ✓
22,786 in / 675 out
2 turns · $ 0.080
Callback -> async/await refactor
benchmarks
41,486 in / 673 out
3 turns · $ 0.102
20,314 in / 165 out
1 turns · $ 0.031 ✓
45,433 in / 357 out
3 turns · $ 0.114
Split a monolith into microservices
benchmarks
20,826 in / 1,575 out
2 turns · $ 0.083
20,550 in / 963 out
1 turns · $ 0.052 ✓
22,807 in / 962 out
2 turns · $ 0.087
PR security review
benchmarks
20,897 in / 1,129 out
2 turns · $ 0.072
19,937 in / 364 out
1 turns · $ 0.037 ✓
22,874 in / 605 out
2 turns · $ 0.079
Multi-stage Dockerfile
benchmarks
42,003 in / 1,561 out
3 turns · $ 0.128
40,412 in / 845 out
2 turns · $ 0.094
22,812 in / 703 out
2 turns · $ 0.081 ✓
Postgres counter race condition
benchmarks
20,827 in / 1,061 out
2 turns · $ 0.070
19,904 in / 480 out
1 turns · $ 0.037 ✓
22,803 in / 530 out
2 turns · $ 0.077
React error boundary component
benchmarks
104,394 in / 2,327 out
6 turns · $ 0.193
89,687 in / 1,096 out
4 turns · $ 0.166 ✓
91,630 in / 2,035 out
5 turns · $ 0.187
Why re-render on parent update?
eval
20,772 in / 738 out
2 turns · $ 0.062
19,870 in / 429 out
1 turns · $ 0.036 ✓
22,752 in / 349 out
2 turns · $ 0.072
Explain connection pooling
eval
20,758 in / 1,267 out
2 turns · $ 0.075
19,863 in / 721 out
1 turns · $ 0.044 ✓
22,737 in / 233 out
2 turns · $ 0.069
TCP vs UDP
eval
20,764 in / 868 out
2 turns · $ 0.065
19,867 in / 369 out
1 turns · $ 0.034 ✓
22,743 in / 269 out
2 turns · $ 0.070
Node.js memory leak
eval
20,779 in / 1,285 out
2 turns · $ 0.076
19,874 in / 713 out
1 turns · $ 0.043 ✓
22,757 in / 693 out
2 turns · $ 0.081
SQL EXPLAIN
eval
20,769 in / 900 out
2 turns · $ 0.066
19,871 in / 472 out
1 turns · $ 0.037 ✓
22,749 in / 497 out
2 turns · $ 0.076
Hash table collisions
eval
20,762 in / 871 out
2 turns · $ 0.065
19,865 in / 309 out
1 turns · $ 0.033 ✓
22,743 in / 470 out
2 turns · $ 0.075
CORS errors
eval
20,770 in / 876 out
2 turns · $ 0.065
19,870 in / 603 out
1 turns · $ 0.042 ✓
22,751 in / 539 out
2 turns · $ 0.077
Debounce a search input
eval
20,772 in / 752 out
2 turns · $ 0.062
19,869 in / 291 out
1 turns · $ 0.033 ✓
22,751 in / 332 out
2 turns · $ 0.071
git rebase vs merge (eval)
eval
20,764 in / 775 out
2 turns · $ 0.063
19,866 in / 437 out
1 turns · $ 0.037 ✓
22,745 in / 403 out
2 turns · $ 0.073
Queue vs topic
eval
20,773 in / 909 out
2 turns · $ 0.066
19,872 in / 484 out
1 turns · $ 0.037 ✓
22,753 in / 344 out
2 turns · $ 0.072
Average, all 20
29,232 in / 1,279 out
2.38 turns · $0.089
24,766 in / 690 out
1.2 turns · $0.053 ✓
27,385 in / 729 out
2.2 turns · $0.088

marks the cheapest arm on that exact prompt. LemonCrow wins 19 of 20. Full per-suite cost/turn breakdown in the flagship page.

The true story

Same prompts, same model, same 5-rep methodology as every other number on this site -- baseline, LemonCrow, caveman, one run. Raw data, all 20 prompts, every arm →