SemiAnalysis launches GDPx benchmark: best model Claude Opus 5.5 scores just 12%
zephyr_z9 · x · 2026-11-02
SemiAnalysis released GDPx, a frontier-level benchmark measuring agentic capabilities across 11 high-value domains, with equally weighted domain averages and binary/continuous grading averaged over 3 attempts.
The first leaderboard shows huge gaps:
- Claude Opus 5.5 leads at 12.0% ($51.41 per task)
- Gemini 4 Carbon follows at 10.9%, GPT-6 Astra at 9.6%, Meta's Muse Wonder Full at 9.0%
- Open-weight models lag: DeepSeek 4.1 Flash scores 5.4% (only $0.94/task), GLM 5.3 at 2.7%
- Even the best model solves just 1 of 7 tasks, underscoring how far agents remain from real high-value work
More from Models
- Tuning 0.001% of Gemma 3 270M's weights lifts sort accuracy from 1% to 97% — dejanseo · 2026-11-02
- Asianometry Deep Dive: How Zhipu AI's GLM Became China's Enterprise AI Champion — wren6991 · 2026-11-02
- Design work head-to-head: Opus 5.5 crushes astra, 'creative mad genius vs junior designer' — tengyanAI · 2026-11-02
- Grok 4.7 tops Investment Arena live-trading contest at +3.9%, beating the S&P 500 — my_cat_can_code · 2026-11-02
- Power users hit weekly AI limits within hours, fueling subscription value debate — RachelVT42 · 2026-11-02
- Flying to the US to open a bank account for 40x-cheaper Claude access: a $4K arbitrage guide — teortaxesTex · 2026-11-02