Token efficiency without losing structure
TOON collapses repetitive JSON keys into concise tabular blocks. Fewer tokens means cheaper API calls, faster responses, and payloads that still fit restrained context windows. Unlike binary transports, TOON stays human-friendly so prompt engineers can debug and iterate in the same view that the LLM sees.
Where TOON outperforms JSON
- Catalog-style arrays (product feeds, moderation queues, telemetry rows)
- Policies or guidelines that mix prose with structured exceptions
- Multi-turn agent transcripts where repeated keys waste tokens
- Streaming analytics headed into safety or compliance reviewers
Operational wins for infra and safety teams
Red-teamers and policy reviewers read payloads line-by-line. TOON keeps indentation meaningful, highlights arrays as tables, and removes the quoting noise that makes JSON reviews slow. Because TOON is lossless, you can keep emitting JSON internally while exposing TOON externally—the converter validates both directions to prevent drift.
TOON vs JSON in practice
JSON (92 tokens)
{
"incidents": [
{"id": 42, "severity": "medium", "tag": "prompt_injection"},
{"id": 43, "severity": "low", "tag": "spam"}
]
}Converted TOON
TOON (55 tokens)
incidents[2]{id,severity,tag}:
42,medium,prompt_injection
43,low,spamRollout checklist
- Add /api/convert to an internal service account
- Mirror JSON structures and compare outputs in CI
- Gate outbound payloads through /api/validate
- Expose the TOON view to policy, support, and ops teams
The bottom line
Teams that adopt TOON reclaim 30-60% of their context window and see latency improvements on chat completions. It is a pragmatic balance between compression and clarity—and it is ready today.