# cubic

> cubic is an AI code reviewer for GitHub. It reviews pull requests to find bugs, security issues, and regressions, can fix what it finds with coding agents, and brings the same review to local workflows through a CLI and an MCP server for coding agents.

## Bug Bench

cubic also publishes Bug Bench, a public leaderboard of AI coding models ranked by relative bug rate: counted bugs per AI-attributed line in code reviewed by cubic, compared with the rate across all AI-attributed lines (1.00 = average, lower is better). Data is CC BY 4.0; cite as "cubic Bug Bench, <date>, https://www.cubic.dev/bugbench".

As of Oct 22, 2026, GPT-6 Luna had the lowest relative bug rate on Bug Bench (0.47, 53% fewer counted bugs than average), among 7 ranked AI coding models.

- [Bug Bench (Markdown)](https://www.cubic.dev/bugbench.md): The full leaderboard, best model by language, bug type, and coding agent, methodology, and citation in one file.
- [Bug Bench](https://www.cubic.dev/bugbench): Live leaderboard of AI coding models ranked by relative bug rate, updated daily.
- [AI coding statistics](https://www.cubic.dev/bugbench/stats): Bug Bench numbers as citable sentences, each linked to the page that shows its data.
- [Data (JSON)](https://www.cubic.dev/bugbench/data.json): Every published rate, rank, and tie in the current snapshot.
- [Data (CSV)](https://www.cubic.dev/bugbench/data.csv): The same rankings in long format.
- [Change feed (Atom)](https://www.cubic.dev/bugbench/feed.xml): Ranking changes, new models, and monthly editions.
- [Models](https://www.cubic.dev/bugbench/models): One page per model with its rank, trend, and coding-agent breakdown.
- [Anthropic models](https://www.cubic.dev/bugbench/providers/anthropic): Anthropic models on Bug Bench ranked by relative bug rate, with unranked and launch-week models listed separately.
- [OpenAI models](https://www.cubic.dev/bugbench/providers/openai): OpenAI models on Bug Bench ranked by relative bug rate, with unranked and launch-week models listed separately.
- [Compare models](https://www.cubic.dev/bugbench/compare): Head-to-head bug rates for any two ranked models.
- [Best model by language](https://www.cubic.dev/bugbench/languages): Rankings within TypeScript, Python, Go, and other languages.
- [Bugs by type](https://www.cubic.dev/bugbench/bugs): Rankings for security, logic, performance, and other bug types.
- [Coding agents](https://www.cubic.dev/bugbench/harnesses): Rankings within Claude Code, Codex, Cursor, and other harnesses.
- [Bug rate vs price](https://www.cubic.dev/bugbench/price): Relative bug rate against list price per 1M tokens.
- [Trends](https://www.cubic.dev/bugbench/trends): Weekly relative bug rates, including whether any model is getting worse.
- [New model launches](https://www.cubic.dev/bugbench/launches): Models in their launch week and how they compare with their predecessors.
- [Monthly editions](https://www.cubic.dev/bugbench/editions): Frozen monthly snapshots for citing a fixed ranking.

## Bug Bench models

- [GPT-6 Luna](https://www.cubic.dev/bugbench/models/gpt-6-luna): #1 of 7 ranked models, relative bug rate 0.47 (53% fewer counted bugs than average), as of Oct 22, 2026.
- [GPT-6.1 Sol](https://www.cubic.dev/bugbench/models/gpt-6-1-sol): #2 of 7 ranked models, relative bug rate 0.98 (about the average rate of counted bugs), statistically tied with Claude Sonnet 5.5, Claude Fable 5.1, and GPT-6 Astra, as of Oct 22, 2026.
- [Claude Sonnet 5.5](https://www.cubic.dev/bugbench/models/claude-sonnet-5-5): #3 of 7 ranked models, relative bug rate 1.06 (6% more counted bugs than average), statistically tied with GPT-6.1 Sol, Claude Fable 5.1, and GPT-6 Astra, as of Oct 22, 2026.
- [Claude Fable 5.1](https://www.cubic.dev/bugbench/models/claude-fable-5-1): #4 of 7 ranked models, relative bug rate 1.07 (7% more counted bugs than average), statistically tied with GPT-6.1 Sol, Claude Sonnet 5.5, and GPT-6 Astra, as of Oct 22, 2026.
- [GPT-6 Astra](https://www.cubic.dev/bugbench/models/gpt-6-astra): #5 of 7 ranked models, relative bug rate 1.09 (9% more counted bugs than average), statistically tied with GPT-6.1 Sol, Claude Sonnet 5.5, and Claude Fable 5.1, as of Oct 22, 2026.
- [Claude Opus 5](https://www.cubic.dev/bugbench/models/claude-opus-5): #6 of 7 ranked models, relative bug rate 1.30 (30% more counted bugs than average), as of Oct 22, 2026.
- [Claude Opus 5.5](https://www.cubic.dev/bugbench/models/claude-opus-5-5): #7 of 7 ranked models, relative bug rate 1.39 (39% more counted bugs than average), as of Oct 22, 2026.

## Code review standards

- [Code review standards](https://www.cubic.dev/standards): Code review rules grouped by language and framework, each shown with code it flags and code that passes.
- [React code review standards](https://www.cubic.dev/standards/react): 12 React review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Next.js code review standards](https://www.cubic.dev/standards/nextjs): 12 Next.js review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [TypeScript code review standards](https://www.cubic.dev/standards/typescript): 11 TypeScript review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Node.js code review standards](https://www.cubic.dev/standards/nodejs): 12 Node.js review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Python code review standards](https://www.cubic.dev/standards/python): 12 Python review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Django code review standards](https://www.cubic.dev/standards/django): 12 Django review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Ruby on Rails code review standards](https://www.cubic.dev/standards/rails): 12 Ruby on Rails review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Go code review standards](https://www.cubic.dev/standards/go): 11 Go review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Spring Boot code review standards](https://www.cubic.dev/standards/spring-boot): 12 Spring Boot review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [PostgreSQL code review standards](https://www.cubic.dev/standards/postgresql): 11 PostgreSQL review rules with examples, ready to paste into AGENTS.md or run on every pull request with cubic.
- [Code review agents](https://www.cubic.dev/agents): All 63 agents by category, including security, database safety, performance, and reliability.

## Comparisons

- [AI code review: what it is and how it works](https://www.cubic.dev/ai-code-review): A guide to AI code review, with Martian’s Code Review Bench results for the leading tools.
- [Best AI code review tools in 2026, ranked](https://www.cubic.dev/best-ai-code-review-tools): The best AI code review tools in 2026, ranked on Martian’s independent benchmark, with verified prices, Git platforms and who each one fits.
- [cubic vs other AI code reviewers](https://www.cubic.dev/vs): How cubic compares with other AI code review tools, with Martian’s Code Review Bench results.
- [cubic vs CodeRabbit](https://www.cubic.dev/vs/coderabbit): cubic ranks #1 on Martian’s Code Review Bench and CodeRabbit #5. Compare them on codebase context, review approach, Git platforms, and pricing.
- [cubic vs Cursor Bugbot](https://www.cubic.dev/vs/cursor-bugbot): cubic ranks #1 on Martian’s Code Review Bench and Cursor Bugbot #9. Compare them on team standards, fixes, Git platforms, and pricing.
- [cubic vs Qodo](https://www.cubic.dev/vs/qodo): cubic ranks #1 on Martian’s Code Review Bench and Qodo #7. Compare them on deployment, Git platforms, IDE support, and pricing.
- [cubic vs GitHub Copilot code review](https://www.cubic.dev/vs/github-copilot): cubic ranks #1 on Martian’s Code Review Bench and GitHub Copilot code review #2. Compare them on setup, codebase context, team standards, fixes, and pricing.
- [cubic vs Codacy](https://www.cubic.dev/vs/codacy): Compare cubic and Codacy on AI review, static analysis, org-wide quality gates, and pricing.
- [cubic vs Sourcery](https://www.cubic.dev/vs/sourcery): Compare cubic and Sourcery on codebase context, Git platforms, security scanning, and pricing.

## Docs

- [cubic docs for LLMs](https://docs.cubic.dev/llms.txt): Index of cubic product documentation, including setup, AI review, the CLI, and the MCP server.
- [cubic docs](https://docs.cubic.dev/): Product documentation home.