What "TCO" actually means for an AI code reviewer
Total cost of ownership for an AI code reviewer is not the seat price on the pricing page. It is seat × months + per-review overages + usage-based credits + hosting + integration + compliance remediation, summed over the contract window. Vendors publish the first number aggressively and quietly assume the rest.
For a 50-engineer team the seat line alone ranges from $0 to $129,600 over three years. That is the widest gap in any dev-tool category I have priced in the last decade. Once you add the four real cost inputs, the gap widens further — CodeOtter's $0 holds up against CodeRabbit Pro because CodeOtter's seat price is real $0 (no per-author, no per-review, no overages) and CodeRabbit's $24 is a teaser that becomes $43,200-$129,600 once you flip on the right plan toggle.
This post walks through the four inputs, ports the 3-year TCO table from our competitor battlecard matrix, and shows the math that an engineering lead should run before signing a renewal.
The four real cost inputs
Every AI code reviewer on the market in 2026 falls into one of these four cost shapes. Quote a vendor without naming the shape and you are getting sold, not sized.
- Per-seat SaaS — CodeRabbit, GitHub Copilot Business. You pay
$N × seats × months. The "Pro" or "Business" tier is the realistic target for most teams. - Usage-based credits — Greptile Pro. $1 per credit, no seat fees; a Base review costs 1 credit, Plus 3, Apex 10 (Greptile pricing page, checked 2026-10-31). The bill follows review volume and depth, not headcount, so a release crunch or a switch to deeper reviews raises it without any new seat appearing.
- Per-review caps — GitHub Copilot Business. 200 PR reviews/month per org on the Business tier. Once you exceed it the dashboard either degrades or you upgrade to Enterprise ($39/seat) for unlimited.
- Self-hosted infra — CodeOtter, Qodo PR-Agent OSS. You pay for the box, the electricity, the GPU (optional), and the engineering hours to wire it. The "$0/seat" headline is real but not zero — it is ~$0–$2,400/yr for a Hetzner AX162 or a pair of M2 Pros, plus the model-load delay on the first review after a quiet period when local GGUF models are in use.
If a vendor's pricing page does not name which of these four shapes they use, treat the number as a starting bid. Ask which tier corresponds to a 50-eng team, what the overage is, and whether the per-author count is active authors or provisioned seats. The answer to the third question decides which headcount a per-seat vendor multiplies by, so price your own team both ways before you compare quotes: every engineer on the roster, and only the engineers who actually open PRs.
The 3-year TCO table for a 50-engineer team
Sources: SaaS list prices from each vendor's pricing page, checked 2026-10-04 (Greptile Pro: 2026-10-31); self-hosted infrastructure, Greptile Pro usage and Qodo Merge figures are our estimates. All numbers are USD, list-price, paid annually.
| Tool | Per-seat | Annual (50 eng) | 3-yr TCO | Cost shape |
|---|---|---|---|---|
| CodeOtter (self-host) | $0 | $0 seats | ~$0–$2,400 infra | Self-hosted (Hetzner AX162 or 2× M2 Pro) |
| CodeRabbit Pro | $24/mo | $14,400 | $43,200 | Per-seat SaaS, cloud-hosted, telemetry required |
| CodeRabbit Pro+ / Team | $48/mo | $28,800 | $86,400 | Per-seat SaaS, adds org-wide policy |
| CodeRabbit Advanced | $72/mo | $43,200 | $129,600 | Per-seat SaaS, adds compliance reporting |
| Greptile Pro | $1/credit, no seat fee | $12,000–$36,000 (our estimate) | $36,000–$108,000 (our estimate) | Usage-based credits (pricing page checked 2026-10-31); estimate assumes 1,000–3,000 Base reviews/mo at 1 credit each |
| GitHub Copilot Business | $19/mo | $11,400 | $34,200 | Per-seat SaaS, 200 PR-reviews/mo cap |
| GitHub Copilot Enterprise | $39/mo | $23,400 | $70,200 | Per-seat SaaS, unlimited PR reviews |
| Qodo PR-Agent OSS | $0 | $0 | ~$9,720–$27,720 (our estimate) | Self-hosted OSS, you operate it; assumes a ~$20/mo VPS plus BYOK LLM API at $5–$15 per dev per month (no System 1 scoring) |
| Qodo Merge (managed) | Custom | ~$15–25/seat (our estimate) | varies | Custom plan for 30+ users (book a demo); no published price at 50 seats |
Three observations from the table that the headline seat price hides:
- The cheapest SaaS is GitHub Copilot Business at $34,200/3yr — and it caps you at 200 PR reviews/month. A team of 50 engineers easily burns 200 PR reviews in a week of feature work plus incident response. Crossing that cap forces an upgrade to Copilot Enterprise ($70,200) or a degraded dashboard. Plan for it on day one, not on month seven.
- The most expensive plan is CodeRabbit Advanced at $129,600/3yr. That is the tier with compliance reporting — the tier a regulated buyer is upsold into. There is no "compliance-lite" tier.
- CodeOtter at $0–$2,400 is not a free lunch. Even on a beefy self-hosted box (2× AMD EPYC, 128 GB RAM, RTX 4090 at ~$300/mo, our estimate), the 3-year run-rate is ~$10,800. That is still about 3.2x cheaper than the cheapest SaaS ($34,200 ÷ $10,800) and 12x cheaper than the most expensive ($129,600 ÷ $10,800). The savings fund the box, the GPU, and a celebration dinner.
The third number is the one I lean on in procurement conversations. A regulated team that buys CodeRabbit Advanced for the audit trail is paying $43,200/yr to a vendor whose telemetry we have to renegotiate in a DPA; the same team running CodeOtter on a Hetzner box pays $300/mo and owns the audit log.
What's not in the headline seat price
Five line items the pricing page will not itemize for you. Plan for them or get ambushed at renewal.
- Per-review credits. Greptile Pro has no seat fee, but every review spends credits at $1 each (Base 1, Plus 3, Apex 10; pricing page checked 2026-10-31). The page publishes the rate, not your monthly volume, so the bill follows review volume and depth. A new engineer adds no seat fee; a release crunch or a move to deeper reviews adds credits.
- PR review caps. Copilot Business caps at 200 PR reviews/org/month. Crossing it does not error — the review either silently degrades or you upgrade. Treat the cap as a soft pricing tier.
- Telemetry compliance work. CodeRabbit and Greptile require telemetry; Copilot routes review code through Microsoft's tenant. A regulated buyer (fintech, healthtech, gov) needs a DPA, a code-routing exception, or both. Estimate 8–16 hours of legal/compliance time per vendor at $200–$400/hr loaded. That is $1,600–$6,400 of work before the first review runs.
- Integration hours. Wiring
AGENTS.mdenforcement, branch protection, GitHub App permissions, Slack/Linear pings, and CI hooks is a 1–3 day half-day minimum per tool. At a $150/hr loaded rate for a senior engineer, that is $1,200–$3,600 of integration work — once, but not zero. - Self-host ops. CodeOtter at $0/seat is not free if you do not have ops bandwidth. Budget 4 hours/month of platform maintenance (model upgrades, GPU driver patches, Ollama updates) at $150/hr loaded = $600/month, $7,200 over three years. For a platform team this is in the noise. For a 5-person startup it is not.
For a 50-engineer team buying a SaaS reviewer, line items 1–4 add $3,000–$10,000 to the 3-year TCO before the first PR is reviewed. For a team running CodeOtter self-hosted, line item 5 adds $0–$7,200 depending on ops capacity. Both numbers are real. Neither vendor will quote them on the pricing page.
Where CodeOtter's $0 holds up — and where it bends
Honest corners:
- $0/seat is real. No per-author. No per-PR. No overages. No PR-review cap. The CLI (
tui/codeotter.mjs) and the platform dashboard both run the same dual-engine System 1 + System 2 pipeline — six calibrated 0–100 gauges and nine binary pre-merge gates in one pass — and neither one bills by the call. - $0–$2,400/yr infra is the realistic low end. A Hetzner AX162 (AMD EPYC, 128 GB RAM, no GPU) handles 50-engineer PR throughput for ~$50/month. With a discrete GPU (RTX 4090 or M2 Pro) for local GGUF inference, $200–$300/month covers it.
- The cold start is real. Local runtimes start on demand with the first review and shut down after 15 idle minutes, so the first review after a quiet period pays the model-load cost and back-to-back reviews do not. The P2 offline mode post explains why it exists.
- Ops is not free. A team without platform bandwidth pays in engineering hours, not dollars. Budget honestly.
The $0 headline survives because the four inputs above (per-seat, usage-based, per-review, infra) are honest line items, not bundled seat tax. The dual-engine System 1 + System 2 architecture is the same W3 CLI-in-CI deep-dive and W2 dual-engine architecture post explain; what changes the math is the billing surface.
What to ask a vendor before signing
Eight questions a vendor should answer in writing before you sign a 3-year deal.
- What is the per-seat, per-author, and per-PR breakdown? Show me the formula.
- What happens if I cross the PR-review cap? Is the dashboard degraded or am I billed for overages?
- What is the telemetry default? Can I turn it off, or do I need a DPA?
- Where does my code route during a review? Does it touch a third-party cloud?
- How do I enforce
AGENTS.md/CLAUDE.mdat the merge gate? Is auto-detection supported? - How do I persist team learnings across PRs? Is there a 1-click "Dismiss & remember rule"?
- What is the audit story for compliance? Can I export the review graph?
- What is the cold-start latency on the first review of the day? What model warms up, and is it local or hosted?
A vendor that answers 1–4 honestly and 5–8 with endpoints, not vibes, is a vendor worth a 12-month pilot. A vendor that hides behind "contact sales" for any of them is a vendor you cannot size without spending the engineering hours to size them.
Battlecards and what to read next
The TCO table above is the math; the battlecards are the architecture, the model roster, and the team-rules story for every individual vendor in the comparison.
- CodeRabbit — CodeOtter vs CodeRabbit (2026). The 72× 3-year TCO gap, the
AGENTS.mdenforcement delta, and why CodeRabbit's.coderabbit.yamlis a config layer not a gate. - Greptile — CodeOtter vs Greptile (2026). The repo-graph Q&A Greptile wins on, our 3-year cost estimate for its per-credit billing, and why Enterprise self-host is a sales-call decision.
- Qodo / PR-Agent — CodeOtter vs PR-Agent (Qodo) (2026). The MIT vs AGPL-3.0 license posture, the estimated 14×–60× TCO gap, and the multi-agent Python CLI workflow library PR-Agent ships that CodeOtter does not replicate.
The canonical four-vendor roundup — the single post that puts all three side-by-side and answers "which one wins at 50 engineers and a regulated buyer?" — is forthcoming in W5 alongside the P4 incremental-delta deep-dive, the P3 AGENTS.md-as-a-merge-gate post, and the P1 vibes-vs-rubrics post. Subscribe to the launch list on the homepage and the post will land in your inbox the day it publishes.
The 3-year TCO for a 50-engineer team runs from ~$0–$2,400 (CodeOtter self-hosted) to ~$129,600 (CodeRabbit Advanced). The gap is honest on the SaaS side because the four real cost inputs — seat, usage, review, compliance — all bill on the same surface. The gap is honest on the self-hosted side because the $0 is a real $0 (no per-seat, no per-PR, no telemetry) and the infra is line-itemed above. Run the eight vendor questions, port the math to your team's PR volume, and the procurement conversation gets shorter by an hour.
CodeOtter ships under the GNU Affero General Public License v3.0 (AGPL-3.0) — free to self-host, $0/seat, zero telemetry, single Docker image + PocketBase + Node 22. The CLI is tui/codeotter.mjs, v0.6.0, zero runtime dependencies. The dual-engine System 1 + System 2 pipeline is the same in the platform dashboard, in CI, and in the local CLI.



