PRICING SNAPSHOT 2026-07-10CLAUDE OPUS 4.5 $5.00/$25.00 MTOKCLAUDE SONNET 4.5 $3.00/$15.00 MTOKCLAUDE HAIKU 4.5 $1.00/$5.00 MTOKGPT-5 $1.25/$10.00 MTOKGPT-5 MINI $0.25/$2.00 MTOKGEMINI 3 PRO $2.00/$12.00 MTOKGEMINI 2.5 FLASH $0.30/$2.50 MTOKDEEPSEEK V3.2 $0.28/$0.42 MTOK
▮▮▮TOKENBURN_INDEX

EDITION 2026 · H2

AI TOKEN BURN INDEX 2026

The public ranking of how money actually leaves your account when you use AI. Updated by edition, argued about constantly.

METHODOLOGY: editorial scores based on public per-token pricing and observed usage patterns. This is commentary with arithmetic, not telemetry. If your habit isn't here, it's not innocence — it's a backlog.

MOST EXPENSIVE AI HABITS

1

Re-pasting the entire conversation "for context"

The model already had the context. Now it has it twice, at input prices.

92
burn
2

Agents left on auto-approve overnight

You woke up to 340 tool calls and a refactor nobody asked for.

88
burn
3

A 2,000-token system prompt for a yes/no question

The answer was 'no'. The invoice was not.

84
burn
4

"Rewrite the whole file" instead of asking for a diff

Output tokens cost 4-8x input tokens. Rewrites are output.

79
burn
5

Transcript → summary → summary of the summary

Recursive compression, expansive billing.

74
burn
6

Regenerating 14 times instead of fixing the prompt once

Slot-machine prompting. The house always wins.

71
burn
7

Politeness rituals: please, thank you, sorry to bother you

Courtesy is free between humans. This is not between humans.

63
burn
8

Decorative markdown headers in internal prompts

The model does not care about your emoji dividers. Your CFO might.

55
burn

PROMPT OBESITY SCORE — BY ROLE

Marketing78

Brand voice documents pasted into every single request.

Founders74

Vision statements where instructions should be.

Consultants71

Frameworks. So many frameworks.

Product managers66

User stories about the prompt inside the prompt.

Developers58

Lean prompts, then pastes the entire monorepo.

Data scientists49

Efficient. Suspiciously quiet about their notebook token usage.

AGENT BURN RISK

EXTREME

Multi-agent crew with a 'reflection' step

Five agents discussing. Conclusion: 'it depends'.

EXTREME

Coding agent in a retry loop on a failing test

Attempt #23 looks a lot like attempt #4.

HIGH

Research agent with unbounded web browsing

It read the whole internet. It cited two tweets.

HIGH

Agent re-reading the full repo every task

Context is not RAM. It is billed like a hotel minibar.

ELEVATED

Scheduled agent that mostly reports 'no changes'

Paying daily to be told nothing happened.

MODERATE

Single agent, scoped task, capped iterations

This is what discipline looks like. Boring, isn't it?

WORKFLOWS THAT BURN MORE THAN EXPECTED

WorkflowThe expectationThe reality
Daily 'summarize all Slack channels'A few centsEvery message re-sent as input, every day. Compounding.
AI meeting notes for every meetingCheaper than a scribeThe summary of the standup cost more than the standup.
Agent code review on every commitQuality gateFull-diff context per commit x commits per day x output verbosity.
Generate 50 variants, pick 1Creative exploration49 deleted drafts, billed at output rates.
Chatbot with full history in every turnIt just remembersNothing 'just' remembers. Input tokens remember, per turn.