Model slugs

One canonical slug per model, across every vendor. log.csv's model column holds a slug from this table and nothing else.

The slug

<vendor-prefix>-<vendor's own model id>

Lowercase, hyphens, no spaces, dots normalized to hyphens. Where the vendor's own id already carries the prefix (Anthropic ships claude-*) the slug is that id unchanged.

Using a model that is not listed here? Mint its slug and add the row on that model's first use, before writing the log row. Never mint a second slug for a model that already has one. Never delete a row — old log entries still point at it; mark it retired in the notes instead.

effort lists the reasoning-effort values the model accepts, or none when it has no such setting. Whatever you actually ran with goes in the log's effort column. none in the table means none in the log.

Rows marked ? could not be confirmed against vendor docs. Correct them the next time you use the model.

anthropic

openai

Each Sol/Terra/Luna tier also ships -fast and -pro variants; append the suffix to the slug (openai-gpt-5-6-sol-pro). The GPT-5 / 5.5 / o3 / o4-mini lineage is deprecated (retires late 2026) and deliberately not listed.

cursor

Cursor names Composer only by display name in-product; no public API id string was found. Confirm and correct on first real use.

xai

xAI documents no reasoning-effort parameter on current models; the old grok-3-mini low/high split no longer applies.

moonshot

Kimi K2.7 Code and K2.6 exist by marketing name; their API id strings were not confirmed. Add rows when you use them.

zhipu

The GLM-5.2 row came from a single unconfirmed source that may have been a template page. Verify before trusting it.

deepseek