AI News Feed
Kimi K2: Open-Source Reasoner Beats GPT-5
08 Nov 2025- Alibaba‑backed Moonshot AI released open‑source Kimi K2 Thinking, matching or beating top closed models on agentic benchmarks, chaining hundreds of tool calls; trained under $5M and priced affordably.
08 Nov 2025
Moonshot AI, an Alibaba-backed startup, has released Kimi K2 Thinking, an open-source reasoning model that the newsletter reports matches or beats leading closed models on a range of agentic benchmarks. According to the scraped coverage, Kimi outperformed GPT-5 and Claude 4.5 Sonnet on several agentic tests and hit a new top score of 44.9% on the benchmark called Humanity’s Last Exam. The model also showed notable coding improvements over Moonshot’s version from four months prior, while still trailing the very top coding specialists.
K2 Thinking reportedly chains 200–300 autonomous tool calls to complete tasks and performs strongly on creative writing. Moonshot says the model cost under $5M to train and will be priced well below current frontier offerings — positioning it as a lower-cost, open-source alternative to major closed models. The newsletter frames the release as evidence for Jensen Huang’s recent comment that China is “nanoseconds” behind U.S. AI efforts, suggesting the open-source and Chinese lab landscape is closing in on frontier capabilities.
Source
Moonshot AI, an Alibaba-backed startup, has released Kimi K2 Thinking, an open-source reasoning model that the newsletter reports matches or beats leading closed models on a range of agentic benchmarks. According to the scraped coverage, Kimi outperformed GPT-5 and Claude 4.5 Sonnet on several agentic tests and hit a new top score of 44.9% on the benchmark called Humanity’s Last Exam. The model also showed notable coding improvements over Moonshot’s version from four months prior, while still trailing the very top coding specialists.
K2 Thinking reportedly chains 200–300 autonomous tool calls to complete tasks and performs strongly on creative writing. Moonshot says the model cost under $5M to train and will be priced well below current frontier offerings — positioning it as a lower-cost, open-source alternative to major closed models. The newsletter frames the release as evidence for Jensen Huang’s recent comment that China is “nanoseconds” behind U.S. AI efforts, suggesting the open-source and Chinese lab landscape is closing in on frontier capabilities.
Source
The method
The prompts
Helping entrepreneurs and business owners use AI tools for growth and productivity.
Wawa Llama Ltd GB447340593