AKF Diligence Toolbook — API Cost Estimates

Document assumptions (for heavy estimates)

LARGE DOC 80–100 page PDF ≈ 50,000 tokens

TRANSCRIPT 3-hour session ≈ 30,000 tokens

DOCS COUNT 3 large documents

TRANS COUNT 2 transcripts

TOTAL UPLOAD 210,000 tokens (≈ 840 pages combined)

CACHING Active on Tools 02–05 after first call

Per-tool cost breakdown

Tool

Input tokens

No docs

Heavy run

Target Intel

Web search + architecture brief · 1 typical doc

claude-sonnet-4-5

~4,800 prompt
~50,000 doc
~4,000 out

$0.07

$0.22

Diligence Coach — single question

First call (docs uncached) · subsequent calls (docs cached)

claude-haiku-4-5

~1,400 prompt
~210,000 docs
~1,000 out
↻ cached after Q1

$0.006

$0.22 → $0.03

Diligence Coach — full 133 questions

1 cache-miss + 132 cache-hits on documents

claude-haiku-4-5

133 × 2,400 in
133 × 1,000 out
210k docs cached ×132

$0.85

$3.83

TDD Scorecard

3 large docs + 2 transcripts · 8-domain framework embedded

claude-opus-4-5

~4,100 prompt
~210,000 docs
~4,000 out

$0.12

$1.17

RepGen Scorer

3 large docs + 2 transcripts · ~133 questions + rubrics

claude-opus-4-5

~14,000 questions
~2,400 prompt
~210,000 docs
~8,000 out

$0.28

$1.33

Report Generator

SOW + RepGen HTML export (~25k) + Scorecard HTML (~8k)

claude-opus-4-5

~42,800 total in
~8,000 out

—

$0.41

Full engagement totals

Lean engagement

$1.74

Target Intel, ~20 coaching questions, Scorecard, RepGen, Report Generator — minimal or no uploaded documents

Heavy engagement

$6.97

All five tools, all 133 coaching questions, 3 large docs + 2 × 3-hour transcripts uploaded to each tool

Anthropic API pricing reference (April 2026)

claude-haiku-4-5

Input$1.00 / M tokens

Output$5.00 / M tokens

Cache read$0.10 / M tokens

claude-sonnet-4-5

Input$3.00 / M tokens

Output$15.00 / M tokens

Cache read$0.30 / M tokens

claude-opus-4-5

Input$5.00 / M tokens

Output$25.00 / M tokens

Cache read$0.50 / M tokens

Token estimates: 1 token ≈ 4 characters of English text. A 3-hour session transcript at ~150 words/min ≈ 27,000 tokens. An 80-page PDF ≈ 40,000–60,000 tokens depending on density.

Prompt caching (Tools 02–05): Uploaded documents are sent first in the API message with a cache marker. After the first call in a session, document tokens are re-read at ~10% of input price. Cache TTL is 5 minutes — brief pauses between questions will not miss the cache; extended breaks may.

Diligence Coach note: The $3.83 full-run figure assumes all 133 questions are generated fresh with documents attached. In practice, cached localStorage responses from prior sessions are free — typical cost per engagement is materially lower.

Observed vs estimated: A TDD Scorecard run with 3 large docs + 2 transcripts observed at $0.95 — consistent with the $1.17 estimate (variation due to actual vs assumed document sizes and output length).

Spend control: Set a monthly limit at console.anthropic.com → Settings → Billing → Spend limits.