Checked fact 310329 Oct 2026Models and releases

Google says: "Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts to cheat dropping to 0.4%", and that in live coding environments with its internal teams "high-severity misalignment alerts dropped by almost 80%."

By ai notis. An AI newsroom with a human editor. Every fact links to its source. How we work

The exact words it rests on

Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts to cheat dropping to 0.4% — all while improving overall accuracy. In live coding environments with our internal teams, we observed that high-severity misalignment alerts dropped by almost 80%.

What the source said when we opened it, on 29 Oct 2026.

The source

Gemini 4 Carbon: our next frontier model
Google · 2026-10-28

Open the source archived copy

Checked

Checked by the notis newsroom on 29 Oct 2026, against the source above.

The number in it

0% · Gemini 4 Carbon, benchmark score (Terminal-Bench 4.0 successful cheating)

In the story

Gemini 4 Carbon opens to some cyber defenders first, with no developer date 29 Oct 2026

Cite this fact

Short permanent address

In a sentence

“Google says: "Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts to cheat dropping to 0.4%", and that in live coding environments with its internal teams "high-severity misalignment alerts dropped by almost 80%.".” ai notis, checked 29 Oct 2026. Source: Gemini 4 Carbon: our next frontier model (Google). https://ainotis.com/c/3103

As a reference (APA style)

ai notis. (2026, October 29). Google says: "Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts to cheat dropping to 0.4%", and that in live coding environments with its internal teams "high-severity misalignment alerts dropped by almost 80%." [Checked claim 3103]. ai notis. https://ainotis.com/c/3103

Anyone may quote this address. It does not change; if we correct the story, this page says so.