Parallel Search Capability Leaderboard

Three hundred questions, 100 each from Google’s DeepSearchQA multi-step research set, Humanity’s Last Exam, and Parallel’s own WISER business-research benchmark. Every model answers them with Parallel Search and Extract under a fixed tool budget, with code execution off, and again with no tools to show how much search adds. The score weights DSQA F1, HLE accuracy, and WISER accuracy equally, and a failed task counts as zero. This board shows the 16 highest-scoring of the 49 models Parallel ranks.

Last updated · Source leaderboard checked daily

Main Insights

Results

Parallel Search Capability Leaderboard results, ranked by score (search intelligence), highest first. Last updated .
RankModelProviderScore (Search Intelligence)Cost (USD / task)
1Claude Fable 5.2Anthropic80.1%$2.43
2Claude Opus 5.5Anthropic76.1%$0.90
3Tencent Hy4Tencent75.4%$0.094
4Gemini 4 CarbonGoogle74.2%$0.47
5Kimi K3.1 ProMoonshot AI74.2%$0.27
6Claude Sonnet 5.5Anthropic74.1%$0.80
7Claude Fable 5.1Anthropic73.4%$1.76
8Gemini 4 HydrogenGoogle72.7%$0.18
9Pareto 26.9Unbiased AI72.5%$0.18
10Gemini 4 ArgonGoogle72.2%$0.27
11Muse Wonder 1.0Meta71.3%$0.71
12GPT-6 AstraOpenAI70.8%$0.40
13GPT-6.1 SolOpenAI70.4%$0.13
14Claude Opus 5Anthropic70.0%$1.01
15GPT-6.1 TerraOpenAI69.7%—
16Particle-1Particle69.2%$0.048
17DeepSeek V4.1 Flash (1017)DeepSeek68.9%$0.13
18MiniMax M3.1 FlashMiniMax68.8%$0.044
19Pareto 26.10Unbiased AI68.6%$0.12
20Ember-1Fireworks67.9%$0.34
21GPT-5.6 SolOpenAI67.7%$0.27
22Muse Spark OpenMeta67.2%$0.17