Demo — Code Lexica

Claude Code & Codex sessions.includes 24 sessions from 4 A/B runs

Est. savings (API-rate value)

$3.2K–$4.1K

vs your team's usage before Code Lexica

Code Lexica sessions — actual $9.5KEst. without Code Lexica $12.7K–$13.5K

Daily spend (API-rate value)

Stacked by attribution; the violet band is the estimated spend if no one had used Code Lexica.

Cumulative savings

With Code LexicaWithoutUntagged — sessions from CLI versions before the Code Lexica signal; shrinks as CLIs upgradeEst. spend if no one used Code Lexica

Window spend (API-rate)$13.8K15%all sessions this window · vs previous window= adoption × behavior × unit price

=Adoption · increase ↑

Devs129%

9 using Code Lexica · 3 not yet

×Agent behavior · decrease ↓

×Unit price · decrease ↓

Model mix

Share of messages by model — the shift from expensive to cheaper models.

Without Code Lexica

claude-sonnet-5 · 50%claude-fable-5 · 40%10%

With Code Lexica

claude-sonnet-5 · 70%10%20%

claude-fable-5 −30ppclaude-sonnet-5 +20ppclaude-haiku-4-5 +10pp

Token types

Share of tokens by type — cache reads are ~10× cheaper than fresh input.

Without Code Lexica

Input · 30%18%Cache read · 42%8%

With Code Lexica

Input · 28%15%Cache read · 48%6%

Cache read +5ppOutput −3ppCache write −2ppInput −2pp

Unit economics

Normalized rates — comparable across windows regardless of team size or volume. The equation above explains what moves them.

API-rate value / 1K messages

$216.0420%

64,100 messages this window

With Code Lexica$196.58

Without$275.08

−29% with Code Lexica

API-rate value / active hour

$11.1918%

1238h of active coding time

With Code Lexica$10.21

Without$14.32

−29% with Code Lexica

Cache hit rate

74%9%

$5,120.40 saved by caching

With Code Lexica78%

Without62%

+16pp with Code Lexica

API-rate value / completed session

$8.0923%

1,712 sessions completed their task — Code Lexica sessions complete 15pp more often

With Code Lexica$6.88

Without$11.59

−41% with Code Lexica

Why spend moved vs last period

2026-05-18 → 2026-07-13, attributed one factor at a time in a fixed order — nothing left unexplained. Devs and messages shown for reference; the comparison is message-grained.

Controlled test results

The same task, run with and without Code Lexica, priced identically — the strongest evidence tier on this page.

All A/B tests

−31%

median cost vs baseline

−28%

median tokens vs baseline

completed controlled runs

Controlled runs are always shown; the A/B switch above governs the observational panels.