74.3%
▲ 59.4 pts since 2023
WebArena success rate of the best entry (Deepseek v3.2), early 2026; human baseline 78.2%
20232026
What the chart shows
WebArena gives agents tasks written in plain language, such as finding information or configuring content across several pages of realistic websites, and checks the end state of the site rather than the steps taken. The best success rate rose from 14.9% (gpt-4-0613) in 2023 to 57.1% (Jace.AI) in 2024, 71.2% (ColorBrowserAgent) in 2025 and 74.3% (Deepseek v3.2) in early 2026.
That leaves agents 3.9 percentage points short of the 78.2% human baseline, the smallest gap among the agent benchmarks the AI Index tracks.
Key numbers
| Measure | Value |
|---|---|
| 2023: gpt-4-0613 | 14.9% |
| 2024: Jace.AI | 57.1% |
| 2025: ColorBrowserAgent | 71.2% |
| 2026: Deepseek v3.2 | 74.3% |
| Human baseline | 78.24% |
How to read it
- 2026 covers results to early 2026.
- Entries mix agent systems (Jace.AI, ColorBrowserAgent) and models (gpt-4-0613, Deepseek v3.2), as listed on the leaderboard.
Source: WebArena Leaderboard, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.6.3: WebArena: success rate. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion