SemiAnalysis launches GDPx agentic benchmark across 11 domains; best model, Opus 5.5, scores 12%
SemiAnalysis has launched GDPx, an agentic benchmark spanning 11 high-value domains including chip design, cybersecurity and recursive self-improvement. The best model scores 12%. SemiAnalysis says safety classifiers depressed the overall scores of Opus 5.5 and GPT-6 Astra.
SemiAnalysis has launched GDPx, an agentic capability benchmark covering 11 domains, and the best model scores just 12%. Opus 5.5 is the top all-around model, according to the report by Max Kan, Andrew Megalaa and Dylan Patel, published November 2, 2026.
What GDPx measures
The 11 domains are chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement (RSI), hardware design, materials science, alignment and software engineering. SemiAnalysis intentionally excludes industries such as government, real estate and insurance.
The authors argue that the first trillion dollars of AI revenue will be concentrated in sectors that can deploy frontier intelligence at high ROI. On that basis they say that benchmarks built on share of GDP or the SOC/O*NET occupational taxonomy are a mistake. They state that GDPx aims to be the world's best benchmark for general agentic capabilities.
Results
- Overall: The best model scores 12%. SemiAnalysis says Opus 5.5 and GPT-6 Astra have sandbagged overall scores because of their safety classifiers.
- Excluding cyber, biology and alignment: These are the three domains where safety classifiers sometimes trigger. With them removed, SemiAnalysis says the ranking matches its subjective impressions. OpenAI and Anthropic lead, while SpaceXAI, Meta and open-weight models are meaningfully behind.
- Gemini 4 Carbon: SemiAnalysis calls it a real improvement from Google and the only model in the same ballpark as Opus and Astra. It notes that Carbon is not yet generally available.
- Cyber: Muse Wonder Full is the best available cyber model by a meaningful margin, according to SemiAnalysis.
Per-domain winners:
- Opus 5.5: quant trading, hardware design and chip design
- Gemini 4 Carbon: biology (SemiAnalysis says likely due to weaker safety classifiers) and a narrow win in software engineering
- GPT-6 Astra: RSI (convincingly) and materials science
- Finance and legal: basically every model scores exactly 0
SemiAnalysis also reports meaningful spikiness across all leading models, and similar spikes within single model families. On the cost versus performance Pareto frontier, Luna and DeepSeek compete at the low end, OpenAI holds the middle and Opus wins the high end. Specific cost figures are on the GDPx website rather than in the newsletter text.
Construction and partners
SemiAnalysis built GDPx with 14 specialized data partners, each an RL environment startup:
- Chip design: Aria, Arcverex, Ergodex
- Cybersecurity: Arithmetic
- Biology: Latch Bio
- Finance: Living Context
- Legal: Chakra
- Quant trading: Third Moment
- RSI: Extrapolate
- Hardware design: Beautiful Machines
- Materials science: ValenceForge, Emulated
- Alignment: Fleet
- Software engineering: Datacurve
For some domains, multiple partners wanted exclusivity, and SemiAnalysis ran a selection process to pick the one with the highest quality tasks. Fireworks provided inference for the open-source models, and Modal provided the sandbox infrastructure. The authors say labs pay $5k+ for a single high-quality task, and that tasks must be hard for the right reasons, avoiding graders that require unstated solutions.
Critique of GDPval
SemiAnalysis argues that OpenAI's GDPval, released September 25, 2025, is inadequate for today's frontier models. It remains the most popular general-purpose benchmark reported on major model release cards, according to the authors. They say every GDPval task is precisely specified and non-interactive, with all context supplied in the prompt, and that its finance example lacks a realistic, rich environment. The authors describe GDPval as impressive for its time and say the shortcoming reflects its pre-agent design.
What this means
GDPx's 12% top score shows these tasks sit well beyond current model capability, which is the authors' stated design goal. The finding that every model scores 0 on finance and legal means those domains do not currently separate models. SemiAnalysis's account of safety classifiers lowering Opus 5.5 and GPT-6 Astra scores means the headline ranking and the ranking without cyber, biology and alignment can point to different models for the same use case. The per-domain winners differ across Anthropic, Google and OpenAI, so a single aggregate number conceals which model leads for a given workload.
Related Articles
Gemini 4 Carbon ranks second on SemiAnalysis GDPx at just over $16 per task, behind Claude Opus 5.5
SemiAnalysis's new GDPx benchmark places Google's Gemini 4 Carbon just behind Anthropic's Claude Opus 5.5, at just over $16 per task against about $51. Removing the cybersecurity, biology and alignment domains reshuffles the leaderboard.
None of 135 AI agent submissions pass experts' bar on open national-lab research in 'Into the Unknown'
A Carnegie Mellon-led team built "Into the Unknown," a benchmark of open research questions from US Department of Energy national labs. According to the paper, none of the 135 submissions from nine AI agents met all requirements, and GPT-6 Astra scored best.
METR report: Anthropic's unreleased Model 4 scores 178.6 on AECI, ahead of Opus 5.5
METR's new report discloses three unreleased Anthropic models, Model 2, Model 3 and Model 4. Model 4 scores 178.6 on Anthropic's AECI index, about 3.9 points above Opus 5.5. Anthropic co-founder Sam McCandlish confirmed the report's main findings but said much information was removed.
CloudOpsBench: Claude Fable 5.1 solves 60% of AWS DevOps tasks, GPT-6 Astra 53%
CloudOpsBench is a new open-source benchmark of 45 AWS DevOps tasks. Claude Fable 5.1 solved 60 percent and GPT-6 Astra 53 percent, and no model completed every task. The research team says reasoning, not knowledge, is the main bottleneck.
Comments
Loading...