Zed token usage
Zed tokens bill through your LLM provider or Zed hosted inference — there is no unified “zed tokens” counter in the IDE like Cursor’s dashboard. Queries for zed tokens usually mean finding where assistant spend shows up.
Hub: Tokenminning in Zed.
Where to look
| Source | What it shows |
|---|---|
| Anthropic / OpenAI dashboard | API token usage for BYOK keys |
| Zed account / subscription | Hosted inference allowance if on Zed plan |
| Assistant thread length | Long threads inflate input every turn |
After a heavy week, check whether spend is input-heavy (rules, @ attachments) or output-heavy (frontier model, long replies). That points you to Context vs model tier changes.
Input vs output
| Pattern | Likely cause | Fix |
|---|---|---|
| High input tokens | Rules bloat, broad @ context | Context guide |
| High output tokens | Frontier default, verbose assistant | Mid-tier model; tighter prompts |
| Slow + high tokens | Marathon thread | New thread per task |
Local vs cloud
Local models trade API tokens for GPU time — still meter wall-clock and utilization for fair comparison against cloud.
Tag spans with ide=zed and inference.backend when you ship agent features behind Zed.
Guardrails
- Provider spending caps on API keys
- Separate keys per developer on teams
- Review provider usage after heavy assistant weeks
Related
Last updated on