57.6%
#1 of 8
Vibe Code Bench accuracy of Claude Opus 4.6 (Nonthinking), the highest of the eight models shown; GPT 5.2 scored 53.5%

What the chart shows

Vibe Code Bench tests whether a model can take a prompt and deliver a working web application end to end, rather than assist with code. Claude Opus 4.6 without extended thinking led the eight models shown at 57.6%, ahead of GPT 5.2 and Claude Opus 4.6 with thinking (both 53.5%).

Scores fall to 48% for GPT 5.4 Mini, 37.9% for GPT 5.2 Codex and 31.5% for GLM 5.1, a spread of 26 percentage points. Even the leader falls short on more than four in ten tasks.

Key numbers

MeasureValue
Claude Opus 4.6 (Nonthinking)57.57%
GPT 5.253.50%
Claude Opus 4.6 (Thinking)53.50%
GPT 5.4 Mini47.97%
GPT 5.2 Codex37.91%
GLM 5.131.46%

How to read it

  • The report text describes an earlier state of the leaderboard (a leader at 56.5%, GPT 5.3 Codex at 41.4%, several models below 15%); we use the figure’s data.
  • “Thinking” and “Nonthinking” are the same model with and without extended reasoning.
  • Scores are from Vals.ai’s Vibe Code Bench v1.1 leaderboard; the AI Index shows eight models.
Source: Vals.ai, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.5.3: Vibe Code Bench v1.1: accuracy. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart