Skip to main content
BenchLM
Data

ResearchClawBench

We show this table for reference; we do not rank on it.

Data verified

An end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper.

RADS average (Pass@1) on ResearchClawBench — 2026-06-29 snapshot

We mirror the published rads average (pass@1) view for ResearchClawBench. Qiushi Engine leads the public snapshot at 38.6%, followed by GPT-5.5 (34.9%) and Ligase (34.8%). We do not use these results to rank models overall.

1

Qiushi Engine

InternScience

38.6%
2Closed

GPT-5.5

OpenAI

34.9%
Overall 66.01Context 1M
3

Ligase

InternScience

34.8%

37 modelsAgenticCurrentDisplay onlyUpdated 2026-06-29 snapshot

RADS average (Pass@1) table (37 models)

RADS average (Pass@1) results for ResearchClawBench
RankModel / configurationScoreParameters (B)Open / closed
1Qiushi EngineInternScience
38.6%
Not reportedUnknown
2GPT-5.5OpenAI
34.9%
Not reportedClosed
3LigaseInternScience
34.8%
Not reportedUnknown
4GPT-5.6 SolOpenAI
33.2%
Not reportedClosed
5GPT-5.5OpenAI
32.7%
Not reportedClosed
6GPT-5.5OpenAI
28.8%
Not reportedClosed
7Qwen3.8 MaxAlibaba
28.4%
Not reportedOpen
8Claude Opus 4.8Anthropic
22.8%
Not reportedClosed
9Claude Opus 4.6Anthropic
21.5%
Not reportedClosed
10Claude Opus 4.8Anthropic
21.1%
Not reportedClosed
11Claude Opus 4.7Anthropic
20.7%
Not reportedClosed
12GLM-5.2Z.AI
20.7%
Not reportedOpen
13Claude Opus 4.6Anthropic
19.9%
Not reportedClosed
14MiniMax M3MiniMax
19.8%
Not reportedOpen
15GPT-5.4OpenAI
18.8%
Not reportedClosed
16Qwen3.7 MaxAlibaba
18.7%
Not reportedClosed
17GPT-5.4OpenAI
18.4%
Not reportedClosed
18GLM-5.1Z.AI
18.2%
Not reportedOpen
19Kimi K2.6Moonshot AI
18.0%
Not reportedOpen
20Qwen3.6 PlusAlibaba
18.0%
Not reportedClosed
21Gemini 3.5 FlashGoogle
17.9%
Not reportedClosed
22DeepSeek V4 Pro 0813DeepSeek
17.1%
Not reportedOpen
23GPT-5.5OpenAI
17.0%
Not reportedClosed
24MiMo-V2.5Xiaomi
16.9%
Not reportedClosed
25GPT-5.4OpenAI
16.6%
Not reportedClosed
26GPT-5.4OpenAI
16.3%
Not reportedClosed
27GPT-5.4OpenAI
15.5%
Not reportedClosed
28MiMo-V2-ProXiaomi
15.3%
Not reportedClosed
29GPT-5.4OpenAI
15.3%
Not reportedClosed
30Qwen3.5 397BAlibaba
14.2%
Not reportedOpen
31Kimi K2.5Moonshot AI
14.0%
Not reportedOpen
32GPT-5.4OpenAI
13.6%
Not reportedClosed
33Grok 4.1xAI
13.5%
Not reportedClosed
34Gemini 3.1 ProGoogle
13.3%
Not reportedClosed
35Hy3 PreviewTencent
12.9%
Not reportedOpen
36GPT-5.4OpenAI
12.8%
Not reportedClosed
37Grok 4.3xAI
12.4%
Not reportedClosed

How ResearchClawBench is shown here

BenchLM mirrors the official ResearchClawBench Pass@1 leaderboard snapshot. The source benchmark contains 40 tasks across 10 scientific domains and uses RADS average (Pass@1) as the primary metric.

ResearchClawBench gives agents related literature and raw data, hides the target paper, and grades how much of the scientific result they rediscover. The RADS scale treats 50 as matching the original paper and 70+ as surpassing it.

ResearchClawBench is display only on BenchLM. The rows combine a model, a research harness, execution budget, and long-horizon scientific workflow, so BenchLM does not use them as weighted base-model ranking inputs.

Snapshot

37 agent rows20 mapped ResearchHarness rows40 tasks10 domains6 Pass@5 rows trackedDisplay only
ResearchClawBench dashboardOfficial leaderboard JSONResearchClawBench paperGitHub repositoryModelScope datasetHugging Face dataset

The published ResearchClawBench snapshot places Qiushi Engine first at 38.6%. The third row is 3.8 points behind. The broader top-10 range is 17.5 points, so the table still separates the published systems.

37 models have been evaluated on ResearchClawBench. The benchmark falls in the Agentic category. ResearchClawBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About ResearchClawBench

Year

2026

Tasks

40 tasks across 10 scientific domains

Format

End-to-end autonomous research evaluation with RADS scoring

Difficulty

Scientific research re-discovery

ResearchClawBench grades scientific agents with RADS, a rubric where 50 indicates matching the target paper and 70+ indicates surpassing it. BenchLM mirrors the official Pass@1 leaderboard as display-only because rows reflect a research-agent harness and long-horizon scientific workflow, not a normalized base-model-only comparison.

ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific ResearchPublic benchmark source

Freshness and provenance

Version

ResearchClawBench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does ResearchClawBench measure?

An end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper.

Which model leads the published ResearchClawBench snapshot?

Qiushi Engine currently leads the published ResearchClawBench snapshot with 38.6% rads average (pass@1). BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on ResearchClawBench?

The 2026-06-29 snapshot snapshot contains 37 AI models.

Last updated: 2026-06-29 snapshot · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.