Pay for what
you actually verify.
No per-seat tax and no mystery model routing. Critique runs automated code verification for GitHub pull requests: sandbox checks, build and type validation, and AI review from a transparent monthly pool. Solo is the safety net for fast-moving Cursor-style builders; Pro and Team scale to serious shared review lanes.
Completed runs finish fast, but deep sandbox verification can still burn real model tokens. The tiers give bigger pools, a free evaluation lane, and a credit gate so low-balance installs stop before one more expensive review or fix run starts.
Frontier models
for everyday repo Q&A.
This product surface is separate from paid PR review: Critique Chat does not draw from the monthly review credit pool you see on model tables and plan cards below. Every signed-in user gets conversational search over connected GitHub repos — pick a model, ask in plain language, and iterate without spending those review credits. (Automated review, Remedy execution, and optional indexing still follow their own pricing.)
- Explain an error from CI with live repo search
- Map a module you have not touched before merge
- Sanity-check a design before you open a PR
Paid plans below are for automated PR review depth, specialist agents, and credit pools — not for chatting with your codebase.
MiniMax M3 at M2.7 pricing through June 17, 2026
Open-weight multimodal coding on the review stack — same 1.5-credit floor as MiniMax-M2.7 today, then 3 credits per run after the welcome window. Qwen3.7 Plus stays on its existing review lane; Critique Chat remains Ling and DeepSeek V4 Flash only.
MiniMax-M3
Review lead & specialist
$8/month harness.
OpenRouter or CrofAI at your cost.
Connect an OpenRouter key and/or a CrofAI (nahcrof) key in Settings. Critique switches chat and sandbox-native OpenCode PR review to your provider — OpenRouter at openrouter.ai or Crof at crof.ai/v1. We run the product harness; your provider bills the raw model calls. When both keys are saved, Crof takes precedence.
E2B sandboxes, OpenCode orchestration, streaming chat, GitHub check runs, repository retrieval, logs, usage ledgers, and secure encrypted key handling.
Frontier and aggregator routes: GPT, Claude, transcription, and any model you route through OpenRouter.
Direct OSS hosting — Kimi K2.6, DeepSeek V4, GLM 5.1, MiMo, Greg 1, and more at https://crof.ai/v1 with OpenRouter-shaped ids mapped to Crof slugs.
Heavy teams should not be squeezed into a low-margin bundle or forced to buy an inflated credit pool just to cover vendor token variance. BYOK makes cost ownership explicit.
Review the submitter before
reviewing the diff.
Checkpoint is a separate pre-review gate. It checks contributor trust, PR shape, changed files, commit history, and policy overrides before the expensive review swarm starts. Run it in dry-run, warn, or block mode, then let approved PRs continue into automated review.
Blocked PRs can skip full review so credits are reserved for code that is ready to inspect.
Modeled as its own policy, override, event log, and GitHub check run rather than a marketing-only switch.
Most review bots start at the diff. Critique can gate risky PRs first, then route the accepted ones through a model panel.
One system.
Simple rules.
| PR Size | Changed Files | Multiplier | Example: GLM-5 + 2× GLM-4.7 Flash (~4.5 base cr) |
|---|---|---|---|
| Tiny | 1–5 | 0.75× | ~3 credits |
| Normal | 6–15 | 1.0× | 5 credits |
| Large | 16–30 | 1.5× | ~7 credits |
| Very Large | 30+ | 2.0× | 9 credits |
Model credits are one side of the equation. Estimate how much manual PR review costs your team in reviewer hours and loaded labor — then compare against a plan.
Three tiers.
One philosophy.
Full-repo embeddings
priced as slots first, credits second.
Enable a repo with Perplexity's pplx-embed-v1-4b and keep the warm snapshot current for semantic retrieval. Chat reads the newest warm snapshot, so you get hybrid search over indexed code without rebuilding context on every question.
Cleaner stack.
Student-pocket pricing.
Get the same 500 credits as the entry pool, but in a tightly curated low-cost lane: GLM-5, Kimi K2.6, and MiniMax M3 (welcome-priced like M2.7 through mid-June) up top, with cheap specialist sub-agents underneath — not the full public catalog. Approved student and OSS accounts also get unlimited free repository indexing.
- First month free
- Friend gets 2 months off
- You get month 2 refunded or month 3 free
- Best for open-source repos, clubs, capstones, and hackathon teams, with free full-repo indexing once approved.
Scale with custom volume, SLAs, and deployment options. On‑prem or cloud, with dedicated support and full compliance.
Every model.
Every cost.
These credit floors apply to automated PR review only — lead and specialist model runs that roll up into your monthly review pool on Solo, Pro, or Team. Critique Chat is separate: repo Q&A and exploration in chat use their own free lane for signed-in users and do not spend review credits. Remedy runs and repository indexing are also their own line items where we describe them above and in the FAQ.
Credits are calculated per review unit — approximately 1M input + 150k output tokens for the lead model on a normal PR. This table does not describe Critique Chat pricing; chat stays outside the review credit meter.
Pick your stack.
See the math.
Estimates below are for PR review credit burn, not Critique Chat. Use Lead then Sub-agent on one list (teal vs green), tweak sub count and depth, and see how each plan's pool stretches.
Tap a row to set the lead reviewer, then switch to Sub-agent for the specialist model.
Questions we hear
before you upgrade.
Credits apply to automated review and to Remedy runs inside Critique — not to everyday Critique Chat or BYOA agent execution (Cursor, Claude, Codex), which bill your vendor accounts directly.
One credit is our normalized unit for a standard review slice — roughly 1M input tokens and 150k output tokens on the lead model for a normal-sized PR. Specialist sub-agents add their own credit cost on top, and the depth multiplier scales everything for tiny vs very large changes. A single full review is often about 3–50+ credits depending on models and PR size. Your monthly pool is meant to cover many reviews, not to map one-to-one with a raw “job” counter.
Solo and Pro share the same standard model lineup and the same flexibility: any listed standard model can be your lead reviewer or a specialist sub-agent, interchangeably. Team adds GPT-5.2 Pro, GPT-5.5 Pro, and Claude Opus 4.7 plus a much larger monthly pool. The internal plan keys are standard, pro, and ultra, but the customer-facing tiers are Solo, Pro, and Team.
Upgrades apply right away so you are not stuck on a smaller pool when volume spikes. Downgrades take effect at the start of your next billing cycle so you keep the features you already paid for through the end of the current period.
No. Monthly credits reset at each billing boundary. We surface costs transparently so teams can right-size plans instead of banking opaque rollover balances.
Pro is advertised with a 7-day trial where that offer is active — see the plan cards above for the exact label on your signup path. Solo and Team follow the CTAs shown there as well.
BYOK is the harness plan for teams that want OpenRouter or CrofAI (nahcrof) to bill model tokens directly. Critique charges $8 for orchestration — sandboxes, OpenCode review runs, chat streaming, GitHub checks, repository retrieval, usage ledgers, and encrypted key storage. Save an OpenRouter key and/or a Crof key in Settings; when both exist, Crof takes precedence. Model spend stays on your provider account instead of Critique credits.
No. Critique Chat is conversational Q&A against connected GitHub repos with frontier models, and it does not draw down the same monthly pool that meters automated PR review. You can explore branches, compare approaches, and ask follow-ups in chat without spending review credits. Repository indexing has its own rules (slots and refresh costs) where applicable — chat itself stays outside the review credit meter.
Remedy is not the chat message — it is the coding agent that edits files, runs managed execution steps, and can trigger follow-up verification passes. That work bills through Critique credits because our stack is carrying the model runs and apply loop end-to-end. The conversation that suggested the fix may be in free chat; turning it into real commits, patches, and checks is what crosses into metered automation.
Those tools bill you through their vendors, not through Critique. Cursor Cloud Agents bill Cursor; Claude Managed Agent and chat handoff bill Anthropic; Codex and OpenAI Responses bill OpenAI; Copilot bills Microsoft / GitHub. Critique does not resell or bundle those subscriptions. Save API keys in Settings to queue Cursor, Claude, or Codex from a review run — or export a blueprint and run it entirely inside your own agent or IDE.
Yes — many teams use Critique for review and queue BYOA runs (Cursor, Claude Managed Agent, OpenAI Codex) or hand execution to Copilot and Claude Code via Fix prompt. In that split, Critique credits cover automated review (and optional Remedy only if you choose it inside Critique), while execution spend stays with whichever agent vendor you already pay.
Indexing is priced separately from review credits: included repo slots, one-time overage enablement, and refresh rules are summarized in the embeddings section above. Student and OSS approvals can include unlimited indexing — check that block for the latest copy. Chat reads warm snapshots once a repo is embedded, but the embedding lifecycle is not the same line item as per-PR review credits.
Write to support@critique.sh for billing, accounts, and general product help. We read every message.
Start with the clearest path.
Create an account to use Critique Chat and connect the GitHub App when your team is ready. Upgrade or change plans from settings anytime.