DeepSearchQA

An agentic browsing benchmark where models search the web, gather evidence, and answer list-style questions using browser tools.

20 models

Higher is better.

1Atria Dawn PreviewShanghai Artificial Intelligence Laboratory96.0
2Claude Opus 5Anthropic95.0
3Kimi K3Moonshot AI95.0
4Claude Opus 4.8Anthropic93.1
5Step 3.7 FlashStepFun92.8
6Kimi K2.6Moonshot AI92.5
7Apodex 1.1Apodex92.4
8dots3-note PreviewDots Studio92.1
9Muse Spark 1.3Meta89.4
10Muse Spark 1.1Meta84.9
11Beam 1.0 501B-A23BReflection AI80.5
12Kimi K2.5Moonshot AI77.1
13Muse SparkMeta74.8
14Muse Glimmer 30BMeta74.6
15Claude Opus 4.6Anthropic73.7
16GPT-5.4OpenAI73.6
17Mercury 3Inception71.3
18Gemini 3.1 ProGoogle69.7
19Grok 4.20xAI62.8
20Mercury 2.5Inception34.0

Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric: Search / open / find browser-agent evaluation
Tasks: Agentic browsing and list-answer questions
Authors: Meta AI (2026)
Source: Muse Spark Eval Methodology ↗

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.