Claude Code: how many tokens per task, and what does it cost?
Claude Code uses 24.1M tokens per task and costs $13.30 per task. AgentBurn snapshot 2026-10-27. Every figure carries a provenance label: source-backed, derived, or left empty when no public basis exists.
What it is. Terminal-native coding agent that plans, edits and verifies changes across a whole repo
Claude Code runs as an interactive CLI inside the repository: it maps the codebase, writes a plan, edits multiple files, runs the test suite and iterates until checks pass, then commits and opens a pull request. Sessions keep a 1M-token context warm through prompt caching, so most of the spend is re-reading context on every turn rather than generating output.
| Tokens per task | 24.1M | Source-backed |
|---|---|---|
| Cost per task | $13.30 | Source-backed |
| Success rate | not published | Not published |
| Cost per solved task | not published | — |
| Tool calls per task | not published | — |
| Retry overhead | not published | — |
| Minutes per task | not published | — |
| Estimated runs per day | 30,000 (model estimate) | Model estimate |
Where the tokens go
This is the per-task split AgentBurn records for Claude Code, step by step.
| Step | What happens | Share of tokens | Tool calls |
|---|---|---|---|
| Orient in repo | Maps the repository, reads CLAUDE.md, lists the failing tests | 6% | 9 |
| Plan the change | Writes an ordered plan and picks the files to touch | 7% | 4 |
| Edit files | Applies multi-file edits with diff review | 22% | 58 |
| Run tests | Executes suites, parses failures and stack traces | 18% | 46 |
| Debug and retry | Reproduces each failure, patches and re-runs | 31% | 92 |
| Verify and commit | Reviews the final diff, writes the commit and opens a PR | 16% | 57 |
What each number is based on
AgentBurn labels every field. Source-backed means the cited page states the figure; derived means a formula on this page produced it; not published means no public basis exists and the field is left empty rather than estimated.
| Field | Basis | Note |
|---|---|---|
| tokensPerTask | Source-backed | Artificial Analysis Coding Agent Index v2.0 — Claude Code Sonnet 5.5 (max), 24.1M tokens/task |
| costPerTaskUsd | Source-backed | AA v2.0 — Claude Code Sonnet 5.5 (max), $13.30/task |
| successRatePct | Not published | AA publishes benchmark scores (Index 59, FrontierCode 48%), not a task success rate. |
| runsPerDay | Model estimate | No public source publishes daily run volume; this is a model estimate from pricing tiers and community reports, shown only in estimated-volume context. |
Frequently asked
How many tokens does Claude Code use per task?
24.1M tokens per task, per the AgentBurn snapshot dated 2026-10-27.
What does Claude Code cost per task?
$13.30 per task, based on the source linked in the provenance table.
Is this a vendor list price?
No. It is a per-task figure from a cited source, not a rate card; the two are not comparable.
Sources
- Artificial Analysis Coding Agent Index v1.5 — Claude Code vs Codex (Oct 2026)
- Vendor site: https://claude.com/claude-code
- Repository: https://github.com/anthropics/claude-code
- Open dataset: /data/projects.json
- Methodology: how these numbers are built
Cite this page
AgentBurn (2026-10-27). "Claude Code". https://getagentburn.com/projects/claude-code/
Machine-readable: projects.json · Markdown version