✦ AI WORK RATE CARD
What AI work roughly costs—from a quick answer to an overnight agent team
The headline: everyday AI work usually costs pennies in raw model usage. Deep research, heavy development, and overnight multi-agent work can cost tens to hundreds of dollars.
QUICK & EVERYDAY TIME POWER AI USE: INPUT / OUTPUT OPENAI CLAUDE
─────────────────────────────────────────────────────────────────────────────────────────────────────
Question, draft, or rewrite 10 sec–2m ●●○○○ 1–10K / 0.2–2K <$0.001–$0.005 $0.002–$0.02
One Slack action 1–5m ●●○○○ 8–40K / 1–5K $0.003–$0.014 $0.01–$0.07
Summarize a document or meeting 3–15m ●●●○○ 20–120K / 2–10K $0.06–$0.36 $0.06–$0.34
RESEARCH & SYNTHESIS
─────────────────────────────────────────────────────────────────────────────────────────────────────
Search Slack/Figma + synthesize 10–45m ●●●●○ 30–250K / 3–25K $0.10–$0.80 $0.09–$0.75
Research, data, or shareable doc 15–90m ●●●●○ 50–400K / 5–60K $0.16–$1.52 $0.15–$1.40
Deep, decision-ready synthesis 2–6h ●●●●● 0.8–4M / 0.15–0.8M $8.50–$44 $7.75–$40
SOFTWARE & BUILDING
─────────────────────────────────────────────────────────────────────────────────────────────────────
Small software change 10–60m ●●●●○ 50–250K / 10–50K $0.22–$1.10 $0.20–$1.00
Difficult debugging or code review 30m–3h ●●●●● 150–800K / 25–120K $1.50–$7.60 $1.38–$7.00
Heavy development day 4–10h ●●●●● 1–6M / 0.2–1M $11–$60 $10–$55
First version of an app from zero 8–24h ●●●●● 3–15M / 0.5–2.5M $30–$150 $27.50–$137.50
EXTREME
─────────────────────────────────────────────────────────────────────────────────────────────────────
Overnight team of 4–8 AI agents 8–16h ●●●●● 15–50M / 2.5–8M $150–$490 $137.50–$450
◉ POWER GUIDE
• ●●○○○ Fast + economical — clear, repeatable tasks · GPT-5.6 Luna / Claude Haiku 4.5
• ●●●●○ Strong + versatile — synthesis, judgment, and most professional work · GPT-5.6 Terra / Claude Sonnet 5
• ●●●●● Maximum reasoning — ambiguity, deep work, coding, and autonomous projects · GPT-5.6 Sol / Claude Opus 5
◉ ADDING MORE AI WORKERS
An agent is an AI worker that can take a goal, use tools, and complete a workstream. Parallel agents help when each has a distinct role—such as researcher, builder, critic, or fact-checker.
HEAVY-WORK TEAM TYPICAL TIME OPENAI SOL TEAM CLAUDE OPUS TEAM
────────────────────────────────────────────────────────────────────────
2 agents 2–6h ~$49 ~$45
4 agents 2–8h ~$90 ~$82.50
8 agents 3–12h ~$158 ~$145
HOW TO READ THIS
- Input is everything the AI reads. Output is everything it writes or reasons through.
- K means thousand tokens; M means million tokens. A token is a small piece of text or data.
- Time is elapsed wall-clock time, not human labor saved.
- Costs are raw, standard API-equivalent model charges—not ChatGPT or Claude subscription prices.
- Search, connectors, image generation, hosting, and human time are not included.
Accuracy note: published model prices and the cost arithmetic are current as of August 12, 2026. Task token bands, elapsed times, power scores, and agent multipliers are informed planning estimates—not vendor benchmarks, guarantees, or SLAs. Overnight runs need a clear scope, spending limit, checkpoints, and morning human review.
OpenAI pricing · OpenAI models · Anthropic pricing · Anthropic model selection