Price the five models the savings table was missing

[amaan-ai20]

Checked ZGPU_PRICING against the dashboard model-pricing API (api-dashboard.zerogpu.ai/api/model-pricing/with-model), the authoritative price list behind the published catalog. Five ids it serves had no entry here, so any call routed to one of them fell through to ZGPU_FALLBACK and was costed at glm-5.2's $1.10 / $3.50.

Model	Real rate	What the CLI was using
zlm-v1-moderation-edge	$0.02 / $0.05	$1.10 / $3.50 (fallback)
t5-small	$0.05 / $0.40	$1.10 / $3.50 (fallback)
zlm-v1-multi-iab-classify	$0.05 / $0.40	$1.10 / $3.50 (fallback)
all-minilm-l6-v2	$0.50 / not billed	$1.10 / $3.50 (fallback)
bge-small-en-v1.5	$0.50 / not billed	$1.10 / $3.50 (fallback)

The one that was actually costing accuracy

zlm-v1-moderation-edge has been in the published catalog since the moderation docs landed, and the fallback prices it 55x over on input. Savings are reported as baseline cost minus ZeroGPU cost, so overstating the ZeroGPU rate understates what the user saved. That is the safe direction to be wrong in, but it is still wrong, and this is the one number in the table a user is asked to trust.

The other four are new to the catalog; nothing was misreporting them yet.

Notes

• out: 0 on the embedding models is the real rate, not a placeholder. Embeddings bill input tokens only. There is a comment saying so, to stop a future reader from "fixing" it.
• No CLI command routes to an embedding model yet. They are priced so the table stays a faithful mirror of the catalog, and so the two-way test keeps passing once the docs list them. A zerogpu embed command is a separate conversation; this PR does not add one.
• ZGPU_FALLBACK is unchanged. glm-5.2 at $1.10 / $3.50 is still the priciest published rate, so the "never overstates savings" invariant still holds.

The test transcription now has one documented exception

tests/savings.test.ts pins ZGPU_PRICING to the catalog in both directions, which is what forces this table to move when the catalog does. Its CATALOG constant is updated in step, and now carries an inline note about the one entry that is deliberately not in the published catalog:

zlm-v1-followup-questions-edge was dropped from the catalog when the model was deprecated, but zerogpu generate-followups still routes to it, so the CLI must still price it. Both entries should be removed together when that command goes.

Verification

• All 40 tests pass, including both directions of the catalog pin
• tsc --noEmit clean, eslint . clean
• Prettier reports pre-existing formatting drift in src/lib/savings.ts and tests/savings.test.ts; both files already fail --check on origin/main, so this PR does not reformat them and add noise to the diff

Related

Docs side: zerogpu/docs#23 adds these models to the published catalog. Independent to merge, but they should land together so neither transcription is stale.

🤖 Generated with Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 84c84379-539c-4bb1-9074-6fb486b03884

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}