**GPT-5.2 & Opus 4.5 Crack Math Problem as Nvidia’s $20B Groq Deal Shakes AI Hardware**
AI Model Advancements
GPT-5.2 Pro and Opus 4.5 Collaborate to Solve Erdős Problem #333: GPT-5.2 Pro, working with Opus 4.5, autonomously resolved a previously unsolved Erdős mathematical problem and formalized the proof in Lean 4—though it was later found that the problem had already been solved in an older paper. This highlights both AI’s growing mathematical capabilities and the need for rigorous literature checks in AI-assisted research.
Claude Opus 4.5 Achieves 4.75-Hour Task Horizon, Setting New SOTA: Opus 4.5 demonstrated a 67% improvement over the previous state-of-the-art, completing software engineering tasks equivalent to ~4.75 hours of human effort. This underscores the rapid scaling of AI capabilities in complex, long-horizon tasks.
GLM 4.7 Ranks #2 on Website Arena, Surpassing GLM 4.6: GLM 4.7 secured the second position on the Website Arena benchmark, trailing only Gemini 3 Pro Preview. This marks a 15-place leap from its predecessor, GLM 4.6, reflecting rapid progress in model performance.
AI Models (GPT-OSS-120B & GLM-4.6) Excel in Civilization V Experiments: Researchers ran 1,408 full games of Civilization V using GPT-OSS-120B and GLM-4.6, revealing distinct AI playstyles and slightly superior performance over the in-game AI. The experiment showcases potential for hybrid AI approaches in strategic gaming.
AI Hardware & Industry Moves
Nvidia Acquires Groq’s Assets for $20B in Record AI Chip Deal: Nvidia is purchasing AI chip startup Groq’s assets for approximately $20 billion, the largest deal in the sector to date. The acquisition signals Nvidia’s aggressive push to dominate AI hardware amid industry consolidation.