October Updates: Any OpenRouter Model, Validated — Plus Honest Review Headlines
A customer wrote in this week: he had typed z-ai/glm-5.3 into the OpenRouter box, and nothing told him whether it worked. It showed up as a bare grey chip. A typo got the same chip. He wasn't sure the model was reviewing anything at all. He was right to be unsure — the model was running fine, but the screen gave him no way to know that. Here's what changed, plus the rest of what shipped in October.
Custom OpenRouter models, now checked against OpenRouter
CodeMouse ships a short featured list of open-weight models (Kimi, GLM, DeepSeek, Qwen), but you've always been able to type any OpenRouter model id. From today, that id is checked against OpenRouter's live catalog the moment you add it:
- A real id becomes a proper row — model name, an Enabled badge, and per-million input/output pricing pulled from OpenRouter, so you can compare it with the featured models at a glance.
- A typo is rejected on the spot, with the offending id in the error, instead of being saved silently and failing on your next PR.
- An id that stops existing (providers do retire models) shows a red Invalid badge with a hint to fix or remove it.
If OpenRouter's catalog is briefly unreachable, we accept the id unverified rather than block you — reviews still run.
So if you want GLM 5.3 — same family as GLM 5.2 with cheaper output — add z-ai/glm-5.3, watch it turn into a priced row, and untick 5.2 if you'd rather switch than add. Each selected model reviews independently and votes in the consensus.
Your PR now tells you when nothing can review it
Over the last month, more than a thousand reviews failed for one dull reason: the organisation had no usable AI provider. A key saved but switched off. A provider switched on without a key. OpenRouter enabled with no model selected. Each of those used to fail quietly in the dashboard while the pull request sat there unreviewed.
Now CodeMouse posts one comment on the PR naming the actual gap in plain words — for example, "xAI and OpenRouter are switched on but have no API key; your Anthropic and OpenAI keys are saved but switched off" — with a link straight to your API keys page. It's posted once per PR, not on every push.
Headlines that never contradict the verdict
Two fixes to the summary comment CodeMouse leaves when it has nothing inline to say:
- If a model asked for changes but had no line to anchor the comment on (often because another reviewer's thread is still open), the summary used to say "no blocking issues found" while the dashboard said changes requested. It now says changes requested and includes the model's reasoning.
- If a review was superseded by a newer push mid-run, the old run no longer publishes anything to the PR. You only ever see comments about the commit that's actually there.
OpenRouter reviews that don't bounce on credit checks
OpenRouter pre-checks your balance against a model's maximum possible output before running. Without a cap, that maximum was 131k tokens, and keys that could easily afford a normal review were turned away with a payment error. Requests are now capped at 64k output tokens — comfortably above what a healthy review produces — with one automatic retry at 24k if the first attempt is refused. Only a genuinely empty key still fails.
Smaller things worth knowing
- Deploys no longer interrupt your reviews. A restarting server hands its in-flight reviews back to the queue, and the new server picks them up.
- Reinstalling the GitHub App on an account with review history now works cleanly; the old installation's repos and history fold into the new one.
- The MCP connector returns a verdict.
list_reviewsandget_reviewnow say approve / comment / request changes, not just "completed", so an agent can't mistake a finished review for a passed one.
All of this started with support emails and production logs. If something about CodeMouse annoys you, [email protected] reaches a human who can ship the fix.
FAQ
Which OpenRouter models can I use?
Any model id listed at openrouter.ai/models. Type it into the box under the featured list; if OpenRouter knows it, it appears as a row with its name and pricing. Up to eight models can be selected at once.
What does the "Invalid" badge mean?
The id isn't in OpenRouter's catalog — usually a typo or a retired model. Reviews can't use it, so remove it or correct the id. Your other selected models keep working.
Does adding a model cost extra?
CodeMouse adds no markup. Each model you select reviews the PR on your own OpenRouter key at OpenRouter's listed price, which is shown on the row.
Why did my PR get a "not configured" comment?
Your organisation had no provider able to run a review at that moment — the comment names the exact reason. Fix it on the API keys page, then comment @codemouse review on the PR or push a commit to re-run.