chore: sync models with dashboard API

[amaan-ai20]

Automated model-catalog sync. The dashboard API is the source of truth; every value below was taken from it. Merging publishes v3.11.0 (a minor bump — every sync is) to npm.

Corrected

Model	Field	Was	Now
deepseek-v4.1-flash	input / output per 1M	$0.30 / $1.20	$0.14 / $0.57
qwen3-30b-a3b-fp8	input / output per 1M	$0.05 / $0.30	$0.10 / $0.45
llama-3.1-8b-instruct-fast	input / output per 1M	$0.02 / $0.05	$0.15 / $0.28

ZGPU_FALLBACK is unchanged — glm-5.2 still holds both the highest input and the highest output rate in the catalog, so the falls back to a default ZeroGPU rate equality test still stands.

The values the sample call test hard-codes llama-3.1-8b-instruct-fast's rate in its expected cost, so that assertion was moved to the new rate alongside the table.

Added

Text Generation — full chat surfaces

All four carry a sample_responses_body, so they are routed responses. That route can be widened later if the platform confirms a Chat Completions path for any of them.

Model	Context	Input / output per 1M
glm-5.3-flash	1,048,576	$0.10 / $0.35
gpt-5.6-luna	272,000	$0.20 / $1.20
gpt-4.1-mini	1,047,576	$0.40 / $1.60
gpt-5.4-nano	400,000	$0.20 / $1.25

Each gets a ZGPU_PRICING + test CATALOG entry, a CHAT_MODELS entry, a row in both chat Models tables, and a mention in the DOCUMENTATION §1 reasoning-and-tool-use bullet.

parameters is null for all four, so their notes cells open with a capability phrase from pricing.description instead of a parameter count. glm-5.3-flash has an empty use_cases, so its cell is sourced from pricing.description alone.

Audio — pricing only

These two appeared in the API after this branch was first cut and were added in a follow-up commit. Neither is Text Generation, so per the sync rules they get a ZGPU_PRICING + test CATALOG entry and nothing else — no chat --model entry, no docs table row, no new per-task command.

Model	Task	Params	Input / output per 1M
whisper-tiny	Speech-to-Text	39M	$0 / $0
chatterbox-nano	Text-to-Speech	110M	$0 / $0

Removed

deepseek-v4-flash-0731 — the API no longer returns it. Dropped from ZGPU_PRICING, the test CATALOG and the never overstates savings model list, CHAT_MODELS and the comment above it, its README row / -m example / routing sentence, its DOCUMENTATION row, §1 bullet, routing paragraph and §5 exceptions sentence, and the ADDING_COMMANDS.md Chat-Completions-only list. No command named it, so none was deleted. The Chat Completions routing sentences still name qwen3-30b-a3b-fp8 and glm-5.2, so nothing cascaded to an empty list.

Rebased onto 3.10.1

main moved while this was open (#13, #14). The 3.10.1 patch release collided with this branch's version bump — both sides only moved the version, so the resolution takes main's 3.10.1 and re-runs npm run bump:minor on it, which lands on the same 3.11.0 this branch already targeted. Merged in as a merge commit; nothing else conflicted.

Notes

Ordering choices, where the skill left slack: the four new Text Generation models are inserted in API (dashboard) order, in the same position in every surface — after deepseek-v4.1-flash, which is the last responses entry and where the removed model sat. The two audio models sit above the embedding block, under their own comment, since the existing out: 0 comment there is specifically about embeddings.

Worth a maintainer's eye: whisper-tiny and chatterbox-nano are priced but not reachable from any current command — the endpoint commands (responses, chat_completions, moderations, embeddings) are all text endpoints. Pricing them is what the sync rules call for; wiring an audio endpoint is a separate change.

Verification

• audit-models.py --strict — 0 findings (24 models)
• npm run lint, npm run build, npm test (56 tests) pass on the merged tree
• zerogpu chat --help lists exactly the API's 11 Text Generation models
• grep confirms deepseek-v4-flash-0731 appears nowhere in src/, tests/, docs/ or README.md

🤖 Generated with Claude Code

https://claude.ai/code/session_014jFCNTvLBuLipdPRdTbg6u

[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: Advanced

Run ID: 54336eb5-d9ba-4a36-bf38-4d6c44bf9825

---

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