Every tool page in a search for free AI code review returns the word "free" in the headline. Read the pricing pages underneath and the word stops meaning one thing. "Free" covers four different deals, and the difference decides whether the tool fits your team or quietly stops working.
The gap is method, not price. Each vendor publishes a pricing page that answers "how much?" but the buyer question is "free under what conditions?" Those conditions are what separate a workable setup from a demo. The same reasoning applies to vendor claims about review quality: a tool can say it catches more bugs than anyone and still offer nothing reproducible to check, which is why every best AI review tool ranking worth reading asks for a method before it trusts a number.
A capped tier stays free, up to a limit
Greptile's Starter plan is genuinely free for one active developer with unlimited repositories. The catch is credits, not cash: 50 credits a month, one active developer, and no cheaper seat to start from. A standard review costs one credit and a TREX runtime review costs three, so the free ceiling is roughly 50 standard reviews a month for a solo developer. The page is explicit that credits reset monthly and do not roll over. Beyond the cap, Pro runs $30 per seat per month with 50 credits per seat, and extra credits are $1 each.
This is the most honest use of the word. The tier is usable for a single contributor evaluating the tool against real pull requests. It stops being free the moment a second developer needs reviews, and it never rolls over, so a quiet month followed by a heavy one does not average out. For a team where agents generate the bulk of the code, the per-review credit model carries an extra implication: the more the agents produce, the faster the free cap burns, because every agent PR still costs review time even if it never merges.
A 14-day trial is not a free tier
CodeRabbit's pricing page opens with "Get a free trial" and "All plans include a 14-day free trial, No credit card required." There is no permanent free tier. Every plan entry, from Essentials at $24 a developer a month billed annually to Advanced at $72, is trial-gated. The plans were renamed recently: Essentials is the former Pro, Team is the former Pro Plus, at the same prices, which the FAQ documents.
A trial is a time-boxed evaluation, not a free tier. It is the correct tool for deciding whether the review quality justifies the price. It is not something a team keeps running. The distinction matters because the search result title says "free" and the subscription starts after fourteen days, which a reader skimming headline pricing will not see.
Free means a service is being wound down
Refact's pricing page returned a 404 during this writing and the site carries a banner that "Refact Cloud is shutting down soon." The product is moving to enterprise self-hosting. A free tier attached to a service that is being discontinued is not a free tier a team should adopt, whatever the headline number was.
Refact is the extreme case of a general truth: "free" is a live fact, not a static one. A pricing page can stop being true without any announcement reaching the buyer. A review tool's free tier is a commitment the vendor can change on its own schedule, so the due diligence is not reading the current number but asking how binding it is. Open-source and self-hosted options externalize that risk, which is one reason the self-hosting cost question keeps coming up in reviews of agent-heavy teams. A self-hosted tool can disappear commercially and still keep reviewing your pull requests. A cloud trial cannot.
Free with your own key
Kodus's Community plan is free and unlimited in pull requests and users, with the condition that you bring your own LLM API key and pay your provider directly. The platform itself is open source and the model cost is separate, itemized in a pricing calculator that breaks monthly spend into LLM tokens plus a per-developer license. The Teams plan is $10 a developer a month with tokens priced separately and no markup on API usage.
This is the fourth meaning: a free license that still carries a model bill. The tool is free, the compute is not. For a team already paying an API provider, the incremental cost is the platform fee, which is zero on the community tier. For a team without an API key, "free" includes a step that is not free. The structure also makes the running cost depend on the model you choose, so the same review volume costs different amounts depending on whether you point a frontier model or a cheap one at the PRs.
What the buyer should actually ask
A search for free AI code review tools surfaces vendor landings that all say "free" while selling different arrangements. The useful question is not "is it free" but "what ends the free period." For Greptile it is the fifty-credit monthly cap and the single seat. For CodeRabbit it is the fourteenth day. For Refact it is the cloud shutdown. For Kodus it is the API key you already pay for.
The same logic applies to self-hosting a review tool, where the infrastructure and maintenance line items can outstrip the license fee. In either case, "free" is upstream of the real cost, and the pricing page is where the method lives.
So read the pricing page with the end condition in mind. A capped tier suits a single evaluator. A trial suits a decision with a deadline. A tool whose cloud is closing suits nobody. A bring-your-own-key license suits a team with an API budget and an appetite for maintaining the integration. All four are called free. None of them are the same product, and none of them answer the buyer question until the end condition is stated.
Top comments (0)