# AgentBurn

> AgentBurn is the independent ledger of what AI agents actually cost: tokens per task,
> cost per solved task and framework overhead across 30 production agents, 22 priced models
> and 3 public benchmarks. Data is published openly as JSON with per-project provenance.

AgentBurn (https://getagentburn.com) tracks real-world AI agent deployments rather than chat
models. Every project card carries a source URL and a provenance tier (Verified, Community,
Estimated), and every leaderboard number is recomputed in the browser from the published
dataset: 24h tokens = runs per day x tokens per task. The dataset is a curated seed snapshot;
a telemetry SDK and ingest API exist for measured data, and the live panel shows real events
once agents are instrumented.

## Key pages

- [Agent project library](https://getagentburn.com/projects): 30 agent projects with tokens/task, cost/task, success rate, run volume, model stack and provenance.
- [Token leaderboards](https://getagentburn.com/leaderboard): 24h / 7d / 30d burn boards, cost per solved task, framework overhead and public benchmark sheets.
- [Cost and ROI calculator](https://getagentburn.com/calculator): price one agent task across 22 models, with retries, cache hit rate, success rate and a human baseline.
- [Live agent ingest](https://getagentburn.com/live): simulated replay of the telemetry event shape (clearly labelled).
- [Industry report](https://getagentburn.com/report): the Agent Token Economy report, with a free summary and a data preview.
- [Methodology](https://getagentburn.com/docs/methodology): metric definitions, provenance tiers, formulas, sources, corrections policy and changelog.
- [Telemetry SDK and API](https://getagentburn.com/docs/sdk): zero-dependency JS and Python clients, ingest endpoints and privacy policy.
- [Contact](https://getagentburn.com/contact): corrections, submissions, sponsorship, privacy and security.
- [About AgentBurn](https://getagentburn.com/about): who runs it, editorial policy, independence and contact.
- [Privacy policy](https://getagentburn.com/privacy): what is collected, retention windows and how to request deletion.

## Data

- [projects.json](https://getagentburn.com/data/projects.json): 30 projects, full workflow breakdowns, sources.
- [models.json](https://getagentburn.com/data/models.json): 22 models with list prices per 1M tokens (input, output, cached).
- [benchmarks.json](https://getagentburn.com/data/benchmarks.json): SWE-bench Verified, Terminal-Bench 3.0, AA Coding Agent Index, framework overhead.
- [meta.json](https://getagentburn.com/data/meta.json): site metadata, data mode (seed/live), disclaimers.
- [traffic.json](https://getagentburn.com/data/traffic.json): Cloudflare edge traffic snapshot used by the admin panel.
- [snapshot.json](https://getagentburn.com/data/snapshot.json): SHA-256 fingerprint of the dataset files. The live endpoint is https://getagentburn.com/api/snapshot.

Dataset snapshot: 2026-10-21. 13 of 30 projects are Verified; the rest are Community or
Estimated with a metrics note explaining the basis. Usage volumes are community-reported or
modelled estimates, not audited billing.

## API

- POST /api/telemetry: batched agent events (input/output/cached tokens, tool calls, retries, success).
- GET /api/stats: aggregated telemetry for the last 30 days.
- POST /api/submit: community project submission (human reviewed).
- GET /api/projects: published community projects (merged into the site at runtime).
- GET /api/health: deployment health and binding status.
- POST /api/lead: report summary lead capture.

## Citation

Cite as: AgentBurn, "Agent project library / Token leaderboards", dataset snapshot 2026-10-21,
https://getagentburn.com, accessed YYYY-MM-DD. Keep the provenance tier attached to any number
you reuse. Full citation guidance: https://getagentburn.com/docs/methodology#cite

## Contact

contact@getagentburn.com

## Optional

- [Terms of use](https://getagentburn.com/terms): dataset reuse, citation and sponsorship disclosure.
- [Refund policy](https://getagentburn.com/refund): 14-day report refund, briefing and sponsorship terms.
- [Privacy policy](https://getagentburn.com/privacy): retention windows and deletion requests.
- [Security contact](https://getagentburn.com/.well-known/security.txt): responsible disclosure.
- [ai.txt](https://getagentburn.com/ai.txt): machine policy for AI systems (retrieval yes, training no).
- [Full context](https://getagentburn.com/llms-full.txt): definitions, provenance tiers and citation guidance.