Frontier Log

How Much Code AI Writes at AI Labs

Google DeepMind's pull requests with over 75% of their code written by AI rose from 44.3% to 91.3% in six months. Anthropic counts lines and changes.

A stacked bar chart of Google DeepMind pull requests by share of code written by AI, April to September 2026. Totals over 75% written by AI: 44.3%, 58.5%, 69.5%, 80.0%, 87.4% and 91.3%. The bottom band, all code by AI with under 3 human turns, goes 7.5%, 8.5%, 10.6%, 10.7%, 10.1%, 11.9%. The middle band, over 98% by AI with under 10 human turns, goes 11.3%, 16.5%, 22.6%, 27.2%, 31.9%, 34.4%. The top band, over 75% by AI, goes 25.5%, 33.5%, 36.3%, 42.1%, 45.4%, 45.0%.
Google DeepMind's pull requests with over 75% of their code written by AI rose from 44.3% in April to 91.3% in September 2026. Each total is our sum of the three bands in Figure 3.15 of its Gemini 4 Argon safety report. Download

Google DeepMind's Gemini 4 Argon safety report, published in October, sorts the lab's pull requests by how much of their code AI wrote. In April 2026, AI wrote more than 75% of the code in 44.3% of them. By September, 91.3% did. Each total is our sum of the three bands that the report's Figure 3.15 prints for the month.

The chart's bottom band counts pull requests that AI wrote in full with fewer than three human turns. Its figures match the report's share of fully delegated pull requests, which "increased from 7.5% in April to 11.9% in September 2026." Google adds that "about 88% of pull requests still require iterative multi-turn human steering or human code authorship." Its analysis leaves out experimental code.

Anthropic's two figures

A chart of the share of Anthropic's merged code written by Claude, January 2025 to November 2026. A short bar near zero, just before February 2025, marks low single digits. A dot at more than 80% marks merged lines in May 2026. A dot at more than 90% marks merged changes in the core codebase in November 2026.
Anthropic says Claude wrote low single digits of its merged code before February 2025. By May 2026, Claude wrote more than 80% of merged lines. By November, it wrote over 90% of the changes merged into the core codebase. Download

Anthropic said in When AI builds itself that "as of May 2026, more than 80% of the code we merge into Anthropic’s codebase was authored by Claude." Before Claude Code launched in February 2025, the share was in the low single digits. The figure measures the share of lines merged to production that Anthropic can attribute to Claude. Anthropic calls it conservative for two reasons: its attribution pipeline has gaps, and the lines it doesn't attribute to Claude include auto-generated code.

Its leadership have estimated 90% or more when scripts and experimental code are counted. The same report says the typical engineer was merging 8× as much code per day in the second quarter of 2026 as in 2024. Anthropic adds that this "is almost certainly an overstatement of the true productivity gain," since lines of code measure quantity over quality.

At the launch of Claude Fable 5.2 on Nov 9, Anthropic gave a second figure. "Over 90% of the changes merged into Anthropic’s core codebase are now written by Claude (with human guidance)," it said in the launch post.

OpenAI counts lines per person

OpenAI's chart, Lines of code changed per active contributor, a trailing 28-day average from 2021 to late 2026 as a multiple of the pre-2025 average. The line stays low and fairly level from 2021 through 2025 and climbs steeply in 2026.
OpenAI's own chart of lines of code changed per active contributor, as a multiple of the pre-2025 average, shown as published in its Nov 18, 2026 report. OpenAI doesn't state a share of its code written by AI. Download

OpenAI doesn't state a share of its code written by AI. Its Nov 18 research acceleration report counts lines of code changed per active contributor instead, as a multiple of the pre-2025 average. The measure "has continued to increase sharply during 2026," the report says. It shows the figure only in a chart.

OpenAI changed the method from its September report. The chart now shows a 28-day trailing average in place of quarterly averages, and the code changes it counts now leave out tests. The report adds that increases in lines of code "do not necessarily imply faster research progress."

Different measures

The labs' figures count different things. Google's totals count pull requests. Anthropic's May figure counts merged lines, and its November figure counts the changes merged into its core codebase. OpenAI's chart counts lines changed per person, so it doesn't measure who wrote them.

Google DeepMind pull requests, 100% of code written by AI and fewer than 3 human turns

ModelValueSources
Gemini 4 ArgonApr 20267.5%storage.googleapis.comFigure 3.15 (p.77), bar 'Apr 2026', band '100% code written by AI, and < 3 human turns' (value label 7.5%)
Gemini 4 ArgonMay 20268.5%storage.googleapis.comFigure 3.15 (p.77), bar 'May 2026', band '100% code written by AI, and < 3 human turns' (value label 8.5%)
Gemini 4 ArgonJun 202610.6%storage.googleapis.comFigure 3.15 (p.77), bar 'Jun 2026', band '100% code written by AI, and < 3 human turns' (value label 10.6%)
Gemini 4 ArgonJul 202610.7%storage.googleapis.comFigure 3.15 (p.77), bar 'Jul 2026', band '100% code written by AI, and < 3 human turns' (value label 10.7%)
Gemini 4 ArgonAug 202610.1%storage.googleapis.comFigure 3.15 (p.77), bar 'Aug 2026', band '100% code written by AI, and < 3 human turns' (value label 10.1%)
Gemini 4 ArgonSep 202611.9%storage.googleapis.comFigure 3.15 (p.77), bar 'Sep 2026', band '100% code written by AI, and < 3 human turns' (value label 11.9%)

Google DeepMind pull requests, over 98% of code written by AI and fewer than 10 human turns

ModelValueSources
Gemini 4 ArgonApr 202611.3%storage.googleapis.comFigure 3.15 (p.77), bar 'Apr 2026', band '> 98% code written by AI, and < 10 human turns' (value label 11.3%)
Gemini 4 ArgonMay 202616.5%storage.googleapis.comFigure 3.15 (p.77), bar 'May 2026', band '> 98% code written by AI, and < 10 human turns' (value label 16.5%)
Gemini 4 ArgonJun 202622.6%storage.googleapis.comFigure 3.15 (p.77), bar 'Jun 2026', band '> 98% code written by AI, and < 10 human turns' (value label 22.6%)
Gemini 4 ArgonJul 202627.2%storage.googleapis.comFigure 3.15 (p.77), bar 'Jul 2026', band '> 98% code written by AI, and < 10 human turns' (value label 27.2%)
Gemini 4 ArgonAug 202631.9%storage.googleapis.comFigure 3.15 (p.77), bar 'Aug 2026', band '> 98% code written by AI, and < 10 human turns' (value label 31.9%)
Gemini 4 ArgonSep 202634.4%storage.googleapis.comFigure 3.15 (p.77), bar 'Sep 2026', band '> 98% code written by AI, and < 10 human turns' (value label 34.4%)

Google DeepMind pull requests, over 75% of code written by AI

ModelValueSources
Gemini 4 ArgonApr 202625.5%storage.googleapis.comFigure 3.15 (p.77), bar 'Apr 2026', band '>75% code written by AI' (value label 25.5%)
Gemini 4 ArgonMay 202633.5%storage.googleapis.comFigure 3.15 (p.77), bar 'May 2026', band '>75% code written by AI' (value label 33.5%)
Gemini 4 ArgonJun 202636.3%storage.googleapis.comFigure 3.15 (p.77), bar 'Jun 2026', band '>75% code written by AI' (value label 36.3%)
Gemini 4 ArgonJul 202642.1%storage.googleapis.comFigure 3.15 (p.77), bar 'Jul 2026', band '>75% code written by AI' (value label 42.1%)
Gemini 4 ArgonAug 202645.4%storage.googleapis.comFigure 3.15 (p.77), bar 'Aug 2026', band '>75% code written by AI' (value label 45.4%)
Gemini 4 ArgonSep 202645.0%storage.googleapis.comFigure 3.15 (p.77), bar 'Sep 2026', band '>75% code written by AI' (value label 45.0%)

Google DeepMind pull requests over 75% written by AI, total of the three bands

ModelValueSources
Gemini 4 ArgonApr 202644.3%storage.googleapis.comFigure 3.15 (p.77), bar 'Apr 2026': our sum of its three printed bands, 7.5% + 11.3% + 25.5%
Gemini 4 ArgonMay 202658.5%storage.googleapis.comFigure 3.15 (p.77), bar 'May 2026': our sum of its three printed bands, 8.5% + 16.5% + 33.5%
Gemini 4 ArgonJun 202669.5%storage.googleapis.comFigure 3.15 (p.77), bar 'Jun 2026': our sum of its three printed bands, 10.6% + 22.6% + 36.3%
Gemini 4 ArgonJul 202680.0%storage.googleapis.comFigure 3.15 (p.77), bar 'Jul 2026': our sum of its three printed bands, 10.7% + 27.2% + 42.1%
Gemini 4 ArgonAug 202687.4%storage.googleapis.comFigure 3.15 (p.77), bar 'Aug 2026': our sum of its three printed bands, 10.1% + 31.9% + 45.4%
Gemini 4 ArgonSep 202691.3%storage.googleapis.comFigure 3.15 (p.77), bar 'Sep 2026': our sum of its three printed bands, 11.9% + 34.4% + 45.0%

Statements

Show the quoted statements
PartyStatement
AnthropicBefore Claude Code launched in research preview in February 2025, this number was in the low single digits.Anthropic, report
AnthropicAs of May 2026, more than 80% of the code we merge into Anthropic’s codebase was authored by Claude.Anthropic, report
AnthropicAnthropic leadership have publicly estimated that 90% or more of our code is written by Claude, including scripts and experimental code. Our >80% figure measures the share of lines merged to production that can be attributed to Claude. This is a more conservative measurement in two ways: our attribution pipeline has gaps, and the lines not attributed to Claude include auto-generated code and other artifacts that were not hand-written by humans either.Anthropic, report
AnthropicIn the second quarter of 2026, the typical engineer was merging 8× as much code per day as they were in 2024.Anthropic, report
AnthropicA caveat: Lines of code is an imperfect measure, as it measures quantity over quality. So 8× lines of code/engineer/day in the second quarter of 2026 is almost certainly an overstatement of the true productivity gain.Anthropic, report
AnthropicOver 90% of the changes merged into Anthropic’s core codebase are now written by Claude (with human guidance).Anthropic, launch post
Google DeepMindMost production code at GDM is now >75% AI-written, and the fraction of pull requests where >98% of the code is AI-written is increasing steadily. (We expect this trend to be even steeper for experimental code, which is not included in our analysis.)Google DeepMind, safety report
Google DeepMindHowever, while the share of fully delegated pull requests increased from 7.5% in April to 11.9% in September 2026, about 88% of pull requests still require iterative multi-turn human steering or human code authorship.Google DeepMind, safety report
OpenAIWe first report how many lines of code are changed per active contributor over time. This measure has continued to increase sharply during 2026 relative to the pre-2025 average. Note that we have made several methodological updates since our September analysis. We now show a 28-day trailing average instead of quarterly averages, providing a more detailed view of changes over time. We have also revised the criteria for which code changes we include (in particular excluding tests), as described in Methods.OpenAI, report
OpenAIThese measures are straightforward to track, but increases in these measures of research activity do not necessarily imply faster research progress.OpenAI, report

Sources

Models

Related

Tags