|
2026年7月16日、中国AIスタートアップのMoonshot AI社が新モデルKimi K3を公開すると、X(旧twitter)上では「中国が米国を超えた」「中国はもはや米国の後追いではない」「AIは新しい時代に入った」という類の声がかなり上がりました。 Kimi K3は2.8兆パラメータのMoE(混合エキスパート)モデルで、100万トークンのコンテキストウィンドウを持っていて、コード生成ではArena.aiのフロントエンドコード部門で、前世代のKimi K2.6が18位だったところから一気に首位へ駆け上がり、米国の最先端モデルClaude Fable 5を上回りました。 コーディング、知識、数学、推論など9つの評価を束ねた英Artificial Analysis社の総合指標ではKimi K3は57点で全体4位。Claude Fable 5(60点)、OpenAI社のGPT-5.6 Sol(59点)に及びませんでしたが、Claude Opus 4.8(56点)を1点上回りGemini 3.5 Flash (50点)を7点も上回りました。 中国のKimi K3が米国のフロンティアモデルの一角に食い込んできたことは間違いなさそうです。このKimi K3について、生成AIに深掘りさせましたので、ご参照ください。なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 中国AIが米国を超えたは本当か Kimi K3の実力とその影響 https://exawizards.com/column/ai-trend/news-07-17-2026/ Moonshot AI Unveils Its New Model, Kimi K3 On July 16, 2026, the Chinese AI startup Moonshot AI unveiled its new model, Kimi K3. The announcement triggered widespread discussion on X (formerly Twitter), with many users posting comments such as, “China has surpassed the United States,” “China is no longer simply following the U.S.,” and “AI has entered a new era.” Kimi K3 is a 2.8-trillion-parameter Mixture-of-Experts (MoE) model featuring a 1-million-token context window. In code generation, it made a remarkable leap to first place in the frontend coding category of Arena.ai, where its predecessor, Kimi K2.6, had ranked only 18th. In doing so, it surpassed Claude Fable 5, one of the most advanced frontier AI models developed in the United States. According to the comprehensive benchmark published by the UK-based Artificial Analysis, which aggregates performance across nine evaluation categories—including coding, knowledge, mathematics, and reasoning—Kimi K3 achieved an overall score of 57, ranking fourth worldwide. While it trailed Claude Fable 5 (60) and OpenAI's GPT-5.6 Sol (59), it edged past Claude Opus 4.8 (56) by one point and outperformed Gemini 3.5 Flash (50) by seven points. It now appears increasingly clear that China's Kimi K3 has joined the ranks of the world's frontier AI models, competing directly with leading U.S. systems. Prompted by these developments, I asked generative AI to conduct an in-depth analysis of Kimi K3 from multiple perspectives. I hope you will find the results informative. Please note that the analysis generated by AI is based solely on publicly available information. It may not fully reflect the actual situation and could contain inaccuracies or errors. Please read it with appropriate caution and critical judgment. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document.
0 Comments
Leave a Reply. |
著者萬秀憲 アーカイブ
April 2026
カテゴリー |
RSS Feed