GPT-5.2 Cracks Erdős Problem as Zhipu AI Makes Historic IPO Debut
Major AI Breakthroughs & Research
GPT-5.2 Solves Unsolved Erdős Problem: Mathematician Terence Tao published a detailed analysis of how OpenAI’s GPT-5.2 solved Erdős Problem #728, the first time an LLM resolved a previously unsolved Erdős problem. The solution revealed the model’s advanced reasoning capabilities and corrected a prior misformulation of the problem.
AI Model Benchmarks & Evaluations
PokerBench: Frontier Models Compete in 21,000 Poker Hands: A new benchmark, PokerBench, pits models like GPT-5.2, GPT-5 Mini, Opus/Haiku 4.5, Gemini 3 Pro/Flash, and Grok 4.1 against each other in poker games. The dataset, including reasoning traces and game replays, is open-source for research.
Comprehensive 4-Bit Quantization Benchmark in vLLM: A study compared AWQ, GPTQ, Marlin, GGUF, and BitsandBytes quantization methods on Qwen2.5-32B (H200). Key findings: Marlin led in speed (712 tok/s), BitsandBytes had minimal quality loss without pre-quantized weights, and GGUF excelled in HumanEval despite higher perplexity.
AI Commercialization & Public Markets
Zhipu AI Becomes First LLM Company to IPO: Chinese AI lab Zhipu AI (creator of the GLM model family) listed on the Hong Kong Stock Exchange, marking the first public offering by a major LLM-focused company. The move signals growing investor confidence in AI commercialization.
- Official: Zhipu becomes the world’s first LLM company to go public
- Z.ai (the AI lab behind GLM) has officially IPO'd on the Hong Kong Stock Exchange
Minimax Follows with Hong Kong IPO: AI company Minimax also went public on the Hong Kong Stock Exchange, continuing the trend of AI firms securing capital for AGI development.
AI in Defense & Government
Mistral AI Partners with French Ministry of Armed Forces: French AI leader Mistral AI signed a historic agreement to provide generative AI solutions to all branches of the French military, reinforcing national technological sovereignty in defense.
AI Agentic Workflows & Tooling
Claude Code Gains Agentic Browsing via Comet Integration: A developer built an MCP server enabling Claude Code to use Comet’s browser (via Chrome DevTools Protocol) for web research, screenshots, and interactive tasks—expanding its agentic capabilities.