OpenRouter Batch Code Review: Open-Weight Models at Half Price in Economy Mode

CodeMouse's economy mode just got its third provider. Starting today, reviews that run through OpenRouter — Kimi K3, GLM-5.2, DeepSeek V4 Pro, Qwen3 Coder Next, or any open-weight model you've selected — can run through OpenRouter's new Batch API at roughly 50% of standard per-token pricing, right alongside the Anthropic and OpenAI batch paths that economy mode already supported.

If you picked OpenRouter models because you watch your spend, this closes the loop: the cheapest models in the CodeMouse lineup now also get the batch discount.

What Economy Mode Does

Economy mode is an org-wide toggle in API Keys that trades speed for cost. With it off, every pull request gets a fast review — CodeMouse clones your repository, lets each configured model browse the code, and posts findings within minutes. With it on, reviews are submitted to your providers' discounted batch endpoints instead: they usually land within an hour (worst case, the provider's 24-hour completion window), the reviewer works from the diff and the PR's discussion context rather than browsing the repository, and every token is billed at the provider's batch rate — typically half price.

Since you bring your own API keys, that discount lands on your bill, not ours. CodeMouse doesn't mark up model usage in either mode.

What's New: OpenRouter Joins the Batch Path

Until now, economy mode covered the two providers that had batch APIs: Anthropic (Message Batches) and OpenAI (Batch API). OpenRouter models — often the cheapest reviewers you can configure — sat the round out, which was exactly backwards for the users most likely to flip the toggle.

OpenRouter's Batch API (currently in beta on their side) changed that. It mirrors the discount model of the other two providers — batch requests are typically billed at 50% of the model's standard per-token pricing — and CodeMouse now uses it whenever economy mode is on and you have OpenRouter models selected.

Mechanically, each model you've picked becomes its own batch submission. If you review with Kimi K3 and GLM-5.2, economy mode submits two batches, polls both, and merges the results through the same consensus pipeline as a normal review: findings are deduplicated across models, prior review threads are checked and resolved, and clean PRs still get approved. A single model's batch failing doesn't sink the review — the surviving models publish, and the failure is logged.

One detail we like: OpenRouter reports the actual charged amount for every completed batch. CodeMouse records that number directly on the review, so the cost you see in your dashboard is what OpenRouter billed — not an estimate from a price sheet.

The Math, With Real Numbers

Take Qwen3 Coder Next, the budget pick in our featured lineup, at roughly $0.11 per million input tokens and $0.80 per million output tokens. A typical mid-sized PR review — call it 40k input tokens of diff and context, 3k output tokens of findings — costs about $0.007 at standard rates. Through the batch path, about $0.0035. Run fifty reviews a day and the arithmetic still barely registers — which is the point. With open-weight models plus the batch discount, per-review cost stops being a reason to review fewer PRs.

The same halving applies to the heavier hitters: a Kimi K3 review that would cost a few cents live costs half that batched. Across a team's monthly PR volume, economy mode with OpenRouter models is the cheapest way to keep every pull request reviewed.

When to Use It — and When Not To

Economy mode is a fit when review latency isn't on your critical path: overnight work, long-running feature branches, repositories where the review is a quality gate rather than a conversation. It's the wrong fit when you want feedback while the PR is still open in your editor — batched reviews usually arrive within the hour, but "usually" is doing real work in that sentence, and the provider's completion window technically allows up to 24 hours.

Two more caveats worth knowing. Batched reviews are diff-scoped: the model sees your diff, the PR's existing comments and threads, and the review instructions — but it can't browse the rest of the repository the way a live review can, so cross-file context is shallower. And Gemini sits economy mode out: it's the one configured provider without a supported batch path in CodeMouse today, so Gemini reviews are skipped while the toggle is on. If Gemini is a load-bearing part of your consensus, leave economy mode off — or add a Gemini-family model through OpenRouter, which now gets batched like everything else.

How to Turn It On

1. Open Dashboard → API Keys.
2. Make sure you have an OpenRouter key connected with at least one model selected (or an Anthropic or OpenAI key — any batch-capable provider enables the toggle).
3. Flip Economy mode at the top of the page. The card spells out the tradeoff, and the Gemini card dims to show it's sitting out.

The toggle is org-wide and owner/admin-only. Turn it off at any time to return to fast, repo-browsing reviews — nothing else about your configuration changes.

FAQ

Does the OpenRouter batch discount apply to my key or to CodeMouse's?

Yours. CodeMouse is bring-your-own-key: batch submissions run against your OpenRouter account, so the ~50% batch pricing lands directly on your OpenRouter bill. CodeMouse never marks up model usage.

What happens if one of my selected models fails its batch?

The review continues with the models that succeeded. A failed or expired batch is logged and skipped; only if every configured provider fails does the review itself fail.

Why is Gemini skipped in economy mode?

Economy mode runs exclusively on provider batch APIs, and Gemini is the one CodeMouse provider without a supported batch integration today. Direct Gemini reviews continue to work normally with economy mode off — and Gemini-family models routed through OpenRouter do get batched.

How long do batched reviews take?

Usually under an hour. All three batch providers work on a 24-hour completion window, so occasional outliers can take longer. CodeMouse polls each batch and posts the review the moment results land.