Skip to content

03. Cost and statistics

🎯 Objective

Measure and optimize cost and tokens with opencode stats, understanding what most impacts the bill.

📊 The stats command

opencode stats              # shareable report
opencode stats --days 7     # 7-day window
opencode stats --models     # usage by model
opencode stats --cost       # cost and tokens
opencode stats --json       # for pipelines

💸 What weighs the most

  1. Large context — the biggest cost factor (input + cache)
  2. Model — switch to #low variants for mechanical tasks
  3. Tool outputs — huge logs/builds inflate the history
  4. Warming — real keep-alive requests
  5. Subagents — each child consumes independently

🛠️ Actions that reduce cost

{
  "compaction": { "auto": true },
  "tool_output": { "max_lines": 2000, "max_bytes": 51200 },
  "agents": {
    "ci-reviewer": { "steps": 8 }
  }
}
  • Pin models by role: cheap for mechanical review, strong for architecture
  • Prefer --format json and process it outside the context
  • Limit subagent steps
  • Turn off warming if there is no cache gain

🧮 Estimate before spending

  • Models have a cost (input/output/cache per 1M tokens) — check the catalog
  • Use opencode run --model provider/model#low for repetitive tasks
  • Monitor automation: record opencode stats --cost as an artifact in CI

🔍 Audit

opencode stats --days 30 --models
opencode stats --cost --json | jq

For teams, the Console adds budgets and a leaderboard.

❓ Hands-on challenge

  1. Measure the cost of a long refactor with --cost
  2. Enable compaction + tool_output and repeat — compare the numbers
  3. Run the same task with #low and #high and compare cost vs. quality
  4. Export --json and build a small report with jq

📚 Official source

Next step: Lesson 04 — Console and enterprise · Back to module