Practical notes on token cost.
Benchmarks and guides from the team building The Distillery. Measured claims, reproducible numbers.
AI Coding Assistant Total Cost of Ownership: The Real Numbers for 2026
The TCO framework: seat fees vs usage costs vs hidden overages. side by side comparison of Claude Code, Cursor, Copilot, and Windsurf at light, moderate, and heavy usage.
Context Engineering for Claude Code: The Ultimate 2026 Guide
How context accumulates in a Claude Code session, why attention distribution matters, when to run /compact, and how to architect projects to stay lean, with real token growth curves.
Prompt Caching for Claude: The Complete 2026 Guide
Updated for the 2026 TTL change. How cache hits and misses are priced, what to cache, the 5-minute window strategy, and where caching stops paying off.
Archive
15 postsThe Complete CLAUDE.md Guide: Best Practices, Templates, and Token Cost in 2026
A definitive 4,000-word guide to CLAUDE.md, cascade loading, token economics, subdirectory patterns, session-notes handoff, and templates for solo devs, teams, and monorepos.
The True Cost of Claude Code for Teams: 2026 Operational Guide
Plan comparison, per dev cost distribution, the Opus 4.6 migration (67% price drop), and the operational habits that separate $80/dev teams from $250/dev teams.
Claude Code Cost Optimization Guide: 30-60% Cost Reduction, Measured
Claude Code costs drop 30-60% with context distillation — 40% measured on a reproducible standard-session benchmark. The methodology, the math, and the one-minute test.
Claude Code Token Usage: Why Each Session Costs More Than the Last
Turn 1 sends 1,000 tokens. Turn 20 sends 50,000+. Why every Claude Code session costs more than the last, and the math behind the curve.
Cursor vs Claude Code (2026): real cost breakdown
Cursor Pro: $20/mo flat. Claude Code at 4+ sessions/day: $300-500/mo. Break-even is 1-2 sessions/day, here is when each is actually cheaper.
Copilot vs Claude Code (2026): $120 vs $4,800/yr TCO
Copilot: $120-228/year flat. Claude Code: $540-4,800/year metered. 12-month TCO compared at light, moderate, and heavy use, verdict by user type.
DISTILLERY_BYPASS: The One Env Var That Gives You Full Control
DISTILLERY_BYPASS turns off all optimization with no warning. The #1 reason developers see no savings. How to check, unset, and verify.
How The Distillery Reduces Claude Code Bills: A Technical Deep Dive
Local Fastify proxy intercepts Claude Code, runs five distillers, forwards smaller payload to Anthropic. 30-60% cost cut, source on GitHub.
How to Benchmark Your Claude Code Token Savings
Measure your Claude Code token savings in under a minute. The benchmark corpus, the two numbers that matter, and how to read the result.
How to Read Your Claude Code Bill: What You're Actually Paying For
80-90% of your Claude Code bill is input tokens, not output. What each line on the Anthropic invoice actually means and where the cost goes.
How to Reduce Claude Code Costs for Your Team (Without Changing Anything)
A 5-dev team on Claude Code burns $2,400/month in API tokens. The exact math and how to cut it 30-60% with no workflow change.
The Same File Read 4 Times: How Claude Code Compounds Token Costs
Claude re-reads package.json 4x per session. A 3,000-token file becomes 12,000 tokens billed, then compounds further. Why, and how to fix it.
What Is Context Engineering? (Why It's the Biggest Driver of Your AI Bill)
80-90% of your Claude Code bill is context you didn't need to send. Context engineering is how you fix that. Four levels, only one is hands off.
Why Claude Code Is So Expensive
Why Claude Code costs $10-50/day: every API turn re-sends the full session context. Where the money goes and how to cut it 30-60% at the proxy.
Why LLMs Waste Tokens (And What You Can Do About It)
Your LLM has no memory between requests, so every API call re-transmits everything. Why the waste compounds and the three practical fixes.