cubic
Try cubic for free
Updated

AI coding statistics

Answers about AI-written code from the pull requests cubic reviewed Jun 22 to Oct 11, 2026, refreshed every day. How we measure

Key statistics

  • What is the most used AI model for coding right now?

    Claude Opus 5.5

    60.5%of AI-written code cubic reviewed in the week of Oct 5, 2026

    Claude Opus 5.5 wrote 60.5% of the AI-written code reviewed by cubic in the week of Oct 5, 2026, more than any other model.

    Source: How widely Claude Opus 5.5 is usedLink to this statistic

  • Does the most used AI model write the fewest bugs?

    No, #7 of 7

    1.5050% more bugs per line than average

    Claude Opus 5.51.50
    GPT-5.6 Sol0.65

    Claude Opus 5.5, the most used AI model (60.5% of the AI-written code reviewed by cubic in the week of Oct 5, 2026), ranks #7 of 7 on Bug Bench with a relative bug rate of 1.50; GPT-5.6 Sol has the lowest (0.65).

    Source: Claude Opus 5.5 bug rateLink to this statistic

  • Which AI model writes the fewest bugs?

    GPT-5.6 Sol

    0.6535% fewer bugs per line than average

    GPT-5.6 Sol had the lowest relative bug rate on Bug Bench (0.65, 35% fewer counted bugs than average), among 7 ranked AI coding models.

    Source: GPT-5.6 Sol bug rateLink to this statistic

  • What is the most used AI coding agent?

    Claude Code

    78.1%of AI-written code cubic reviewed in the week of Oct 5, 2026

    Claude Code wrote 78.1% of the AI-written code reviewed by cubic in the week of Oct 5, 2026, more than any other coding agent.

    Source: How widely Claude Code is usedLink to this statistic

Models

Bug rate: 1.00 is average, lower is better

  • Which AI model writes the most bugs?

    Claude Opus 5.5

    1.5050% more bugs per line than average

    Claude Opus 5.5 had the highest relative bug rate on Bug Bench (1.50, 50% more counted bugs than average), among 7 ranked AI coding models.

    Source: Claude Opus 5.5 bug rateLink to this statistic

  • How big is the gap between the best and worst AI model?

    57% fewer bugs per line

    0.65 vs 1.50GPT-5.6 Sol vs Claude Opus 5.5

    GPT-5.6 Sol0.65
    Claude Opus 5.51.50

    GPT-5.6 Sol had 57% fewer counted bugs per line than Claude Opus 5.5 (0.65 vs 1.50).

    Source: GPT-5.6 Sol vs Claude Opus 5.5Link to this statistic

  • Do newer AI models write fewer bugs than the ones before them?

    No

    2 of 2 write more bugs per line than the model before them

    Of the 2 AI models ranked alongside the model before them in their family, both have higher relative bug rates: Claude Opus 5.5 (1.50 vs 1.33 for Claude Opus 5) and GPT-6.1 Sol (1.23 vs 0.65 for GPT-5.6 Sol).

    Source: Claude Opus 5.5 vs Claude Opus 5Link to this statistic

  • Is any AI model getting worse?

    Yes, 2 of 4

    4 models have a weekly trend

    Of the 4 AI coding models with a weekly trend, 2 have rising bug rates: GPT-5.6 Sol (+0.47) and Claude Opus 5 (+0.13).

    Source: Weekly bug rate trendsLink to this statistic

Coding agents

Compared with the average across all models

  • Which AI model writes the fewest bugs in Claude Code?

    Claude Fable 5.1 and Claude Sonnet 5.5

    1.09 to 1.24Too close to call

    Claude Fable 5.11.09
    Claude Sonnet 5.51.24

    Claude Fable 5.1 had the lowest relative bug rate in Claude Code (1.09, 9% more counted bugs than the all-model average), statistically tied with Claude Sonnet 5.5, among 4 ranked models.

    Source: Best AI model for Claude CodeLink to this statistic

  • Which AI model writes the fewest bugs in Codex?

    GPT-5.6 Sol

    0.6238% fewer bugs per line than average

    GPT-5.6 Sol had the lowest relative bug rate in Codex (0.62, 38% fewer counted bugs than the all-model average), among 3 ranked models.

    Source: Best AI model for CodexLink to this statistic

  • Which AI model writes the fewest bugs in Cursor?

    Claude Opus 5 and GPT-5.6 Sol

    1.28 to 1.52Too close to call

    Claude Opus 51.28
    GPT-5.6 Sol1.52

    Claude Opus 5 had the lowest relative bug rate in Cursor (1.28, 28% more counted bugs than the all-model average), statistically tied with GPT-5.6 Sol, among 2 ranked models.

    Source: Best AI model for CursorLink to this statistic

Adoption

Share of AI-written code reviewed by cubic

  • Which AI company has the largest share of AI-written code?

    Anthropic

    73.1%of AI-written code cubic reviewed in the week of Oct 5, 2026, by its models on Bug Bench

    Anthropic's models on Bug Bench wrote 73.1% of the AI-written code reviewed by cubic in the week of Oct 5, 2026, more than any other company's models, and OpenAI's models wrote 17.2%.

    Source: Adoption on the leaderboardLink to this statistic

  • Which AI model gained the most share?

    Claude Opus 5.5

    60.5%of AI-written code, new since the week of Sep 21, 2026

    Claude Opus 5.5, first seen in the week of Sep 21, 2026, wrote 60.5% of the AI-written code reviewed by cubic in the week of Oct 5, 2026, the largest gain of any model since the week of Jun 22, 2026.

    Source: How widely Claude Opus 5.5 is usedLink to this statistic

  • Which AI model lost the most share?

    Claude Opus 4.8

    0.2%of AI-written code, down from 60.9% in the week of Jun 22, 2026

    Claude Opus 4.8's share of the AI-written code reviewed by cubic fell from 60.9% in the week of Jun 22, 2026 to 0.2% in the week of Oct 5, 2026, the largest drop of any model.

    Source: How widely Claude Opus 4.8 is usedLink to this statistic

Languages

Compared with the average in each language

  • Which AI model writes the fewest bugs in TypeScript?

    GPT-5.6 Sol and Claude Sonnet 5.5

    0.68 to 0.78Too close to call

    GPT-5.6 Sol0.68
    Claude Sonnet 5.50.78

    GPT-5.6 Sol had the lowest relative bug rate in TypeScript code (0.68, 32% fewer counted bugs than the TypeScript average), statistically tied with Claude Sonnet 5.5, among 7 ranked models.

    Source: Best AI model for TypeScript codeLink to this statistic

  • Which AI model writes the fewest bugs in JavaScript?

    GPT-6.1 Sol and GPT-5.6 Sol

    0.72 to 0.96Too close to call

    GPT-6.1 Sol0.72
    GPT-5.6 Sol0.96

    GPT-6.1 Sol had the lowest relative bug rate in JavaScript code (0.72, 28% fewer counted bugs than the JavaScript average), statistically tied with GPT-5.6 Sol, among 7 ranked models.

    Source: Best AI model for JavaScript codeLink to this statistic

  • Which AI model writes the fewest bugs in Python?

    GPT-6.1 Sol, GPT-5.6 Sol, and Claude Fable 5.1

    0.70 to 0.84Too close to call

    GPT-6.1 Sol0.70
    GPT-5.6 Sol0.71
    Claude Fable 5.10.84

    GPT-6.1 Sol had the lowest relative bug rate in Python code (0.70, 30% fewer counted bugs than the Python average), statistically tied with GPT-5.6 Sol and Claude Fable 5.1, among 7 ranked models.

    Source: Best AI model for Python codeLink to this statistic

  • Which AI model writes the fewest bugs in Go?

    GPT-5.6 Sol and GPT-6 Astra

    0.81 to 0.94Too close to call

    GPT-5.6 Sol0.81
    GPT-6 Astra0.94

    GPT-5.6 Sol had the lowest relative bug rate in Go code (0.81, 19% fewer counted bugs than the Go average), statistically tied with GPT-6 Astra, among 6 ranked models.

    Source: Best AI model for Go codeLink to this statistic

  • Which AI model writes the fewest bugs in SQL?

    GPT-5.6 Sol

    0.3565% fewer bugs per line than the SQL average

    GPT-5.6 Sol had the lowest relative bug rate in SQL code (0.35, 65% fewer counted bugs than the SQL average), among 6 ranked models.

    Source: Best AI model for SQL codeLink to this statistic

  • Which AI model writes the fewest bugs in Rust?

    GPT-6 Astra and GPT-5.6 Sol

    0.47 to 0.68Too close to call

    GPT-6 Astra0.47
    GPT-5.6 Sol0.68

    GPT-6 Astra had the lowest relative bug rate in Rust code (0.47, 53% fewer counted bugs than the Rust average), statistically tied with GPT-5.6 Sol, among 5 ranked models.

    Source: Best AI model for Rust codeLink to this statistic

  • Which AI model writes the fewest bugs in Ruby?

    Claude Fable 5.1

    0.4357% fewer bugs per line than the Ruby average

    Claude Fable 5.1 had the lowest relative bug rate in Ruby code (0.43, 57% fewer counted bugs than the Ruby average), among 5 ranked models.

    Source: Best AI model for Ruby codeLink to this statistic

  • Which AI model writes the fewest bugs in Shell?

    GPT-5.6 Sol, Claude Fable 5.1, and GPT-6 Astra

    0.58 to 1.16Too close to call

    GPT-5.6 Sol0.58
    Claude Fable 5.10.93
    GPT-6 Astra1.16

    GPT-5.6 Sol had the lowest relative bug rate in Shell code (0.58, 42% fewer counted bugs than the Shell average), statistically tied with Claude Fable 5.1 and GPT-6 Astra, among 5 ranked models.

    Source: Best AI model for Shell codeLink to this statistic

  • Which AI model writes the fewest bugs in Vue?

    Claude Fable 5.1 and GPT-6 Astra

    0.35 to 0.91Too close to call

    Claude Fable 5.10.35
    GPT-6 Astra0.91

    Claude Fable 5.1 had the lowest relative bug rate in Vue code (0.35, 65% fewer counted bugs than the Vue average), statistically tied with GPT-6 Astra, among 5 ranked models.

    Source: Best AI model for Vue codeLink to this statistic

  • Which AI model writes the fewest bugs in Swift?

    Claude Opus 5.5 and GPT-6 Astra

    0.56 to 0.74Too close to call

    Claude Opus 5.50.56
    GPT-6 Astra0.74

    Claude Opus 5.5 had the lowest relative bug rate in Swift code (0.56, 44% fewer counted bugs than the Swift average), statistically tied with GPT-6 Astra, among 4 ranked models.

    Source: Best AI model for Swift codeLink to this statistic

  • Which AI model writes the fewest bugs in Svelte?

    GPT-5.6 Sol, Claude Opus 5.5, and GPT-6 Astra

    0.39 to 0.94Too close to call

    GPT-5.6 Sol0.39
    Claude Opus 5.50.77
    GPT-6 Astra0.94

    GPT-5.6 Sol had the lowest relative bug rate in Svelte code (0.39, 61% fewer counted bugs than the Svelte average), statistically tied with Claude Opus 5.5 and GPT-6 Astra, among 4 ranked models.

    Source: Best AI model for Svelte codeLink to this statistic

  • Which AI model writes the fewest bugs in Kotlin?

    Claude Opus 5 and Claude Opus 5.5

    0.68 to 0.86Too close to call

    Claude Opus 50.68
    Claude Opus 5.50.86

    Claude Opus 5 had the lowest relative bug rate in Kotlin code (0.68, 32% fewer counted bugs than the Kotlin average), statistically tied with Claude Opus 5.5, among 3 ranked models.

    Source: Best AI model for Kotlin codeLink to this statistic

  • Which AI model writes the fewest bugs in C#?

    GPT-5.6 Sol

    0.3368% fewer bugs per line than the C# average

    GPT-5.6 Sol had the lowest relative bug rate in C# code (0.33, 68% fewer counted bugs than the C# average), among 3 ranked models.

    Source: Best AI model for C# codeLink to this statistic

  • Which AI model writes the fewest bugs in PHP?

    Claude Opus 5

    0.6733% fewer bugs per line than the PHP average

    Claude Opus 5 had the lowest relative bug rate in PHP code (0.67, 33% fewer counted bugs than the PHP average), among 3 ranked models.

    Source: Best AI model for PHP codeLink to this statistic

  • Which AI model writes the fewest bugs in Terraform?

    GPT-5.6 Sol and Claude Opus 5.5

    0.64 to 0.79Too close to call

    GPT-5.6 Sol0.64
    Claude Opus 5.50.79

    GPT-5.6 Sol had the lowest relative bug rate in Terraform code (0.64, 36% fewer counted bugs than the Terraform average), statistically tied with Claude Opus 5.5, among 3 ranked models.

    Source: Best AI model for Terraform codeLink to this statistic

Bug types

Compared with the average for each kind of bug

  • Which AI model writes the most security and privacy bugs?

    Claude Opus 5.5, Claude Sonnet 5.5, and Claude Opus 5

    1.22 to 1.50Too close to call

    Claude Sonnet 5.51.22
    Claude Opus 51.37
    Claude Opus 5.51.50

    Claude Opus 5.5 had the highest rate of security and privacy bugs among 7 ranked models, 50% above the all-model rate (1.50), statistically tied with Claude Sonnet 5.5 and Claude Opus 5.

    Source: Which AI model writes the most security and privacy bugs?Link to this statistic

  • Which AI model writes the most logic bugs?

    Claude Opus 5 and Claude Opus 5.5

    1.32 to 1.36Too close to call

    Claude Opus 5.51.32
    Claude Opus 51.36

    Claude Opus 5 had the highest rate of logic bugs among 7 ranked models, 37% above the all-model rate (1.36), statistically tied with Claude Opus 5.5.

    Source: Which AI model writes the most logic bugs?Link to this statistic

  • Which AI model writes the most testing and documentation issues?

    Claude Opus 5.5, Claude Sonnet 5.5, and GPT-6.1 Sol

    1.74 to 2.03Too close to call

    Claude Sonnet 5.51.74
    GPT-6.1 Sol1.83
    Claude Opus 5.52.03

    Claude Opus 5.5 had the highest rate of testing and documentation issues among 7 ranked models, 103% above the all-model rate (2.03), statistically tied with Claude Sonnet 5.5 and GPT-6.1 Sol.

    Source: Which AI model writes the most testing and documentation issues?Link to this statistic

  • Which AI model writes the most error-handling bugs?

    Claude Opus 5, GPT-6.1 Sol, and Claude Opus 5.5

    1.17 to 1.33Too close to call

    GPT-6.1 Sol1.17
    Claude Opus 5.51.32
    Claude Opus 51.33

    Claude Opus 5 had the highest rate of error-handling bugs among 6 ranked models, 34% above the all-model rate (1.33), statistically tied with GPT-6.1 Sol and Claude Opus 5.5.

    Source: Which AI model writes the most error-handling bugs?Link to this statistic

  • Which AI model writes the most performance bugs?

    GPT-6.1 Sol, GPT-6 Astra, and Claude Opus 5.5

    1.38 to 1.96Too close to call

    GPT-6 Astra1.38
    Claude Opus 5.51.55
    GPT-6.1 Sol1.96

    GPT-6.1 Sol had the highest rate of performance bugs among 6 ranked models, 96% above the all-model rate (1.96), statistically tied with GPT-6 Astra and Claude Opus 5.5.

    Source: Which AI model writes the most performance bugs?Link to this statistic

  • Which AI model writes the most config and dependency issues?

    GPT-6.1 Sol, Claude Opus 5, and Claude Opus 5.5

    1.36 to 1.93Too close to call

    Claude Opus 51.36
    Claude Opus 5.51.68
    GPT-6.1 Sol1.93

    GPT-6.1 Sol had the highest rate of config and dependency issues among 6 ranked models, 93% above the all-model rate (1.93), statistically tied with Claude Opus 5 and Claude Opus 5.5.

    Source: Which AI model writes the most config and dependency issues?Link to this statistic

  • Which AI model writes the most type and compile errors?

    Claude Opus 5.5 and 3 others

    1.03 to 1.35Too close to call

    GPT-6 Astra1.03
    Claude Fable 5.11.17
    Claude Opus 51.33
    Claude Opus 5.51.35

    Claude Opus 5.5 had the highest rate of type and compile errors among 5 ranked models, 35% above the all-model rate (1.35), statistically tied with GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.

    Source: Which AI model writes the most type and compile errors?Link to this statistic

  • Which AI model writes the most business-logic bugs?

    Claude Opus 5.5

    2.55155% more business-logic bugs per line than average

    Claude Opus 5.5 had the highest rate of business-logic bugs among 5 ranked models, 155% above the all-model rate (2.55).

    Source: Which AI model writes the most business-logic bugs?Link to this statistic

  • Which AI model writes the most accessibility and UX issues?

    Claude Opus 5.5

    1.9494% more accessibility and UX issues per line than average

    Claude Opus 5.5 had the highest rate of accessibility and UX issues among 5 ranked models, 94% above the all-model rate (1.94).

    Source: Which AI model writes the most accessibility and UX issues?Link to this statistic

  • Which AI model writes the most code-quality issues?

    GPT-6 Astra and 3 others

    1.08 to 1.31Too close to call

    Claude Fable 5.11.08
    Claude Opus 5.51.15
    Claude Opus 51.18
    GPT-6 Astra1.31

    GPT-6 Astra had the highest rate of code-quality issues among 5 ranked models, 31% above the all-model rate (1.31), statistically tied with Claude Fable 5.1, Claude Opus 5.5, and Claude Opus 5.

    Source: Which AI model writes the most code-quality issues?Link to this statistic

  • Which AI model writes the most architecture and design issues?

    GPT-6 Astra, Claude Opus 5.5, and Claude Opus 5

    1.28 to 2.05Too close to call

    Claude Opus 5.51.28
    Claude Opus 51.33
    GPT-6 Astra2.05

    GPT-6 Astra had the highest rate of architecture and design issues among 4 ranked models, 105% above the all-model rate (2.05), statistically tied with Claude Opus 5.5 and Claude Opus 5.

    Source: Which AI model writes the most architecture and design issues?Link to this statistic

  • Which AI model writes the most style and convention issues?

    GPT-6 Astra, Claude Opus 5, and Claude Opus 5.5

    1.11 to 1.77Too close to call

    Claude Opus 51.11
    Claude Opus 5.51.36
    GPT-6 Astra1.77

    GPT-6 Astra had the highest rate of style and convention issues among 4 ranked models, 77% above the all-model rate (1.77), statistically tied with Claude Opus 5 and Claude Opus 5.5.

    Source: Which AI model writes the most style and convention issues?Link to this statistic

How to read these numbers

Every answer covers only code written by AI models and coding agents, in pull requests cubic reviewed Jun 22 to Oct 11, 2026. Read the methodology.

Relative bug rate
Counted bugs per line in a model's code, compared with an average. For models and coding agents that is all AI-written code; for a language or a kind of bug it is the average for that language or kind of bug. 1.00 is average: 0.80 means 20% fewer bugs per line, and 1.50 means 50% more.
Counted bugs
Findings from cubic code review on AI-written code that a developer upvoted or an AI coding agent addressed.
Statistically tied
Rates too close to call with the data so far, so neither model is clearly ahead.

For cubic's wider findings on AI coding, read the State of AI coding 2026 report.