Skip to content
Trending storyDeveloping

Google's Gemini 4 Carbon reports state-of-the-art on Terminal-Bench benchmarks

1 article1 sourcesince Oct 30Last article 4h ago ·

Overview

AISummary of 1 article

Google team member Sumit Bagri says the team's contribution to Gemini 4 Carbon has reached state of the art on Terminal-Bench 4.0 and Terminal-Bench-Science.

In his post, Bagri reports 64.8% on Terminal-Bench-Science and says cheating attempts on Terminal-Bench 4.0 fell to 0.4%. He adds that how the model reaches its answers matters as well as the scores.

Written by AI from the articles below · updated Oct 31, 12:20 AM ET

Check the sources:

Latest developmentOct 30, 8:55 PM ET
Gemini 4 Carbon sets SOTA on Terminal-Bench 4.0 and Terminal-Bench-Science

Article timeline

The articles in this story. Times are ET.

Oct 30
  1. Sumit BagriX
    Gemini 4 Carbon sets SOTA on Terminal-Bench 4.0 and Terminal-Bench-Science

    AIA Google team member says their contribution to Gemini 4 Carbon has reached state of the art on Terminal-Bench 4.0 and Terminal-Bench-Science. The post reports 64.8% on Terminal-Bench-Science and cheating attempts reduced to 0.4% on Terminal-Bench 4.0. The author adds that how the model reaches its answers also matters.

    Official Google benchmark table for Gemini 4 Carbon, with Carbon’s column highlighted in blue. Terminal-Bench 4.0 (AA): Carbon 65.2%, Argon 57.1%, GPT-6 Astra 59.1%, Claude Opus 5.5 59.6%, Claude Fable 5.1 52.0%. Terminal-Bench-Science 0.1 (AA): Carbon 64.8%, Argon 52.9%, GPT-6 Astra 63.3%, Claude Opus 5.5 59.0%, Claude Fable 5.1 43.3%. Other rows cover general capabilities, coding, computer use, knowledge work, multimodal understanding, long context, science and math, and cybersecurity.

Heat Trend

Not enough continuous observations to show a trend yet.