SemiAnalysis Launches GDPx Benchmark; All AI Agents Score Below 12 Percent
Current overview (not historical)
SemiAnalysis has launched GDPx, a real-world work benchmark that places AI agents inside simulated knowledge-work environments to measure their performance on practical tasks. The suite includes 206 tasks spanning 11 domains—chip design, biology, quantitative trading, cybersecurity, and law—and every AI agent tested scored no more than 12 percent. The Decoder reported that independent verification of the benchmark's task difficulty and scoring quality remains difficult for outsiders, and noted that safety filtering mechanisms could reshape the final rankings. The report treats the results as unverified by third parties.
Written with AI from the reporting · updated 11/5/2026
Expand the detailed fact route and sources
Story route
A station-by-station route from the first public evidence onward; every station carries its own evidence and uncertainty.
- 1Unclassified fact stationNov 6, 01:13single sourcereport-time onlyrevised
SemiAnalysis 发布 GDPx 基准，所有 AI 代理得分均不超过 12%
Read original source ↗ The DecoderReplay this moment in the explorerEvidence & revisions · 1 source records
- Subject：
- SemiAnalysis
- Action：
- 发布 GDPx 基准并评测 AI 代理
- Object：
- AI 代理 / 大语言模型
- Occurred (fact side)：
- not recorded
- Published (valid side)：
- Nov 6, 00:46
- First seen (known side)：
- Nov 6, 01:13
Fact stationSingle source · one-sided claimRevisedReport-time onlyCopy may have been revised later; time replay is bounded. Layer labels group reading only, not causation between layers.
Uncertainty marks flag where to check the sources; they are not proven causation.
Report timeline
Explore every side through dated source reports.
- The DecoderRelated fact: single sourceFact occurrence time unverifiedSource report revisedSemiAnalysis GDPx benchmark shows no AI agent scores above 12 percent
Report first recorded Nov 6, 01:13
SemiAnalysis has released GDPx, a real-world work benchmark that places AI agents in simulated knowledge-work environments across 11 domains and 206 tasks. No model scored above 12 percent, with Claude Opus 5.5 leading at 12 percent, Gemini 4 Carbon at 10.9 percent at about one-third the cost, and GPT-6 Astra at 9.6 percent. Safety filters and closed task sets also affect the rankings and verifiability.
Read original source ↗
Event activity trend
Not enough continuous observations to chart activity.
Event discussion
Reader opinions are hosted on GitHub Discussions; posting requires a GitHub account. Comments are not verified facts.