$ git diff main -- faq.md
commit a1b2c3d - author: Oliver Wakefield-Smith - verified 21 Jun 2026
14 buyer-side questions on AI code review tools, answered with the source URL in the magenta pin. Bundled Copilot Code Review is free at the seatsrcbut consumes AI Credits per review since June 2026. Per-question detail below.
@@ 14 questions @@priority order
1Is GitHub Copilot Code Review free if I already pay for Copilot?
At the seat level, yes: Copilot Code Review is bundled in Copilot Business ($19/user/mo) and Copilot Enterprise ($39/user/mo). Since 1 June 2026, individual reviews now consume AI Credits, and overage above the included allotment bills at list.
2What is the cheapest AI code reviewer for a 10-developer team?
Qlty at $10/dev/mo and Sourcery / DeepSource at $12/dev/mo are the cheapest published seats. Whether they fit depends on your Git platform and deployment posture. Run the calculator.
3Can AI code review tools run fully on-prem?
Yes, but the list is short: Qodo, Augment Code, Snyk Code, SonarQube Server, Diffblue Cover. Most others are SaaS-only or require an outbound connection even on a self-hosted install.
4Do AI code review tools replace human reviewers or augment them?
Augment. As of mid-2026, every published vendor still treats AI review as a pre-merge filter that a human still signs off on. Vendors that claim 'replaces human review' are exaggerating.
5What is the difference between SAST and AI code review?
SAST (SonarQube, Snyk Code, DeepSource) runs rule-based pattern matching. AI code review (CodeRabbit, Greptile, Claude Code Review) generates prose comments that explain the reasoning. The two stack: SAST catches the known, AI catches the novel.
6Does CodeRabbit work on GitLab and Azure DevOps?
Yes. CodeRabbit supports GitHub, GitLab, Bitbucket and Azure DevOps with first-class PR/MR integration.
7What does Cursor BugBot really cost?
$40 per user per month for the add-on, PLUS Cursor Business at $40 per user per month as a prereq. So the effective entry cost for a buyer not already on Cursor is $80 per seat.
8How does Claude Code Review price?
Base seat at $100/seat/mo on the Team plan, plus pass-through Anthropic API tokens at Sonnet 4.5 list ($3 input, $15 output per 1M tokens). $/PR scales with PR size.
9What is BYOK and which tools support it?
BYOK = bring your own LLM key (Claude / Gemini / OpenAI) so token cost is paid directly to the model vendor. Today: Qodo, Bito, Macroscope, Git AutoReview, and Claude Code Review (which IS the BYOK path).
10How much does an AI review cost in tokens at Claude Sonnet 4.5 list?
Roughly $0.025 per PR at 250 LOC. Inputs run 24 tokens per LOC (PR context inflates), outputs run 2 tokens per LOC.
11Are accuracy benchmarks like OpenSSF or SWE-bench usable for vendor selection?
As directional signal, yes. As a ranking column, no. Numbers move per release; a static snapshot misleads. We point at the leaderboards but do not roll them into a 'score'.
12What changed in AI code review pricing in 2026?
Three big moves. (1) Claude Code Review launched Mar 2026 with $100 seat + token meter. (2) GitHub switched Copilot from flat premium requests to AI Credits on 1 Jun 2026. (3) Macroscope v3 raised the team seat from $18 to $22/dev/mo.
13Does Copilot Code Review work on Azure DevOps?
Yes, on Azure DevOps Repos with the Copilot Business or Enterprise tier. Azure DevOps Server coverage lags GitHub on Azure DevOps Services.
14When is BYOK genuinely cheaper than a managed reviewer?
Above the per-tool breakeven on /byok-breakeven. For most 10-dev teams at 50 PRs/wk, BYOK undercuts managed once monthly PR volume passes the high single-thousand mark. Add ~10% ops overhead before comparing.
@@ cross-links @@related
1/calculator - model your team
2/scorecard - all 19 side by side
3/byok-breakeven - BYOK crossover volumes
4/methodology - how we source pricing