SemiAnalysis launches GDPx benchmark: best frontier model scores just 12%
rajistics · x · 2026-11-03
SemiAnalysis released GDPx, a benchmark spanning the most valuable domains for frontier intelligence: chip design, cybersecurity, bio, finance, legal, quant trading, RSI, hardware design, materials science, alignment and software engineering — deliberately skipping government, real estate and insurance, on the thesis that AI's first trillion dollars of revenue will concentrate in high-ROI sectors.
Key results:
- It's a true frontier-level benchmark: even the best model scores only 12% overall.
- No single model dominates: Opus 5.5 leads hardware design (18.72), GPT-6 Astra tops AI self-improvement (24.89), Gemini 4 Carbon wins software engineering (11.11), and Muse Wonder Full takes cybersecurity (35.19).
- Opus 5.5 and GPT-6 Astra are sandbagged by safety classifiers in cyber/bio/alignment; excluding those domains changes rankings.
- OpenAI and Anthropic dominate the top; SpaceXAI, Meta and open-weight models trail meaningfully.
The takeaway: the best model varies by task, so agents should stay model-agnostic.
Related event: SemiAnalysis Launches GDPx Frontier Benchmark; Best Models Score Just 12%(12 posts)→
More from Models
- Mistral Large 4 tech report: 1T-param MoE, 606GB quantized, runs on four GB300 GPUs — _AndrewZhao · 2026-11-03
- Mistral Large 4 goes open-weight with frontier code-agent and cyber capabilities — _AndrewZhao · 2026-11-03
- Are Chinese models eating the market? thdxr calls it selection bias — yenkel · 2026-11-03
- Altman says OpenAI's internal model that "solved Navier-Stokes" is coming "soon" — haider1 · 2026-11-03
- X trending: DeepSeek V4.1 Flash and GLM 5.3 Flash rumored to lead open-model race (unverified) — itsmnjn · 2026-11-03
- Grok Bots testing private sharing: invite users by email, team-wide shared bot browser — XFreeze · 2026-11-03