Of 16,341 T3 Code installs that ran Google’s Gemini 4 Neon in its first 53 hours, 5,306, or 32%, had run Anthropic’s Opus 5.5 most the week before, according to a chart posted by Adam Holter. Holter, who said T3 Code is where he got the data, captioned it “Neon is stealing users from Opus.”
The same chart shows those 5,306 were 9% of Opus 5.5’s active installs, so most Opus users did not switch in the window. Neon became half or more of the turns for 6,330 of the 16,341 installs, 38.7% by our arithmetic.
The data cover one coding tool and the 53 hours from 18 to 21 November. They show which model users picked, not which one did the work better; for scores, see our AI model benchmarks tracker.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer. It teaches the general idea only and does not discuss this chart.
The key insight: Of 16,341 T3 Code installs that ran Gemini 4 Neon in its first 53 hours, 5,306 (32%) had used Opus 5.5 most the week before, per a chart by Adam Holter; that was 9% of active Opus 5.5 installs, and Neon became the main model for 6,330 installs (38.7%).
Where Neon’s Users Came From
Holter’s chart sorts every install that ran Neon by the model it used most from 10 to 17 November. Opus 5.5 was the largest group, at 5,306 installs.
The next largest, 4,939 installs, had not used T3 Code at all that week. GPT-6.1 Sol users accounted for 1,544, other Gemini models for 1,540 and Fable for 1,016.
Other OpenAI models added 784, other models 676 and other Claude models 536. Together, installs whose main model was a Claude model made up 6,858, or 42%, by our arithmetic.

Share of Each Group That Tried Neon
The chart’s percentages read differently: each one is the share of that group’s installs active after launch that ran Neon, per its footnote.

By that measure 63% of installs that mainly used other Gemini models tried Neon, against 9% for Opus 5.5 and 10% for GPT-6.1 Sol. Large numbers from Opus reflect how many installs ran Opus, not a high switching rate.
Working back from 9%, roughly 59,000 Opus 5.5 installs were active after launch, by our arithmetic; because the 9% is rounded, the true figure is between about 56,000 and 62,000.
How Much They Used It
For 6,330 installs, Neon became half or more of their turns from 18 to 20 November; for 10,011, it stayed under half, the chart shows. Holter counts a Neon turn as one that names the model or uses Antigravity’s default setting, Neon High.
Holter separately called Neon “Google’s best launch in T3 Code” on 21 November. Google priced Neon at $3 per million input tokens and $15 per million output, as we reported, and it led LiveBench this week.
The Structural Read
As we read it, the chart shows Opus 5.5 as the biggest source of Neon’s early users because Opus has the biggest user base in T3 Code, not because Opus users switched at a high rate.
As we read it, a launch-week window measures trial; whether Neon keeps these users is the number that matters for Google and Anthropic.
Adam Holter, on X
“Neon is stealing users from Opus”
Three Implications
FOR ANTHROPIC In this chart, 5,306 Opus 5.5 installs ran Neon, which was 9% of active Opus 5.5 installs.
FOR GOOGLE Neon became half or more of turns for 6,330 of 16,341 installs that tried it in T3 Code, 38.7%.
FOR CODING-TOOL MAKERS The chart counts installs and turns in one tool over 53 hours; it does not measure output quality.
What Is Not Established
The window is 53 hours, from 21:00 UTC on 18 November to 02:00 UTC on 21 November, right after launch, when trying a new model is common. The chart does not show whether these users stay on Neon or return to Opus.
The chart covers users of one coding tool, T3 Code, which has not published the underlying dataset. Anthropic and Google have not commented on the chart.
What This Means
The chart shows Neon drew its largest group of users from Opus 5.5, while about nine in ten active Opus 5.5 installs did not run it in the window.
The Bottom Line
In T3 Code, 32% of Neon’s early users came from Opus 5.5, and 39% made Neon their main model within days, per Holter’s chart. What to watch is whether that share holds after the launch week.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. All install counts and percentages are from the chart Adam Holter posted on X on 22 Nov 2026; he said the data come from T3 Code, which has not published the dataset. The 32%, 38.7%, 42% and the roughly 59,000 Opus installs are our arithmetic from his figures. We first saw the chart via a link shared with our newsroom.
Sources: Adam Holter on X, ‘Neon is stealing users from Opus’, chart of T3 Code installs that ran Gemini 4 Neon, 22 Nov 2026 · Adam Holter on X: ‘T3 Code is just where I got the data from’, 21 Nov 2026 · FourWeekMBA, Gemini 4 Neon Tops LiveBench at 87.4, 22 Nov 2026 · FourWeekMBA, Gemini 4 Neon Costs $3/$15, 18 Nov 2026









