layout | published |
---|---|
default |
false |
- OpenAI שחררה למנויי פרו את Operator - סוכן AI נסיוני המסוגל לגלוש באתרים ולבצע פעולות
- Google שחררה את Gemini Flash thinking 0121, מודל חשיבה משופר. המודל הגיע למקום הראשון ב-Arena chatbots
- Deepseek שחררה כקוד פתוח מודלי חשיבה בשם R1 ו-R1-Zero, שהציג יכולות דומות למודל o1 במגוון תחומים ושבריר מהמחיר. בנוסף שחררה החברה מודלים מזוקקים קטנים יותר שהציגו ביצועים גבוהים מאוד ביחס לגודלם
- Google הציגה נייר מחקר של ארכיקטורה חדשה לבניית מודלי שפה בשם Titans המאפשרת למודלים לשמור זיכרון לטווח קצר וארוך. הארכיקטורה משפרת משמעותית עיבוד של חלון הקשר מוגדל
- Deepseek שחררה כקוד פתוח מודל רב מודאלי מלא בשם Janus Pro 7B הכולל יצירת תמונות וטקסט כאחד.
- עליבאבא חשפה את Qwen2.5-Max, מודל שפה גדול שמנצח מספר דגמים מובילים כמו Deepseek-V3 , GPT-4o , ו-Claude 3.5. בנוסף שחררה החברה כקוד פתוח סדר מודלים בשם Qwen2.5-1M המסוגלים לנתח עד מיליון אסימונים ואת סדרת דגמי ראייה Qwen2.5-VL בשלושה גדלים.
- OpenAI released Operator for Pro subscribers – an experimental AI agent capable of browsing websites and performing actions.
- Google introduced Gemini Flash Thinking 0121, an enhanced reasoning model that secured the top spot in the Arena Chatbots rankings.
- DeepSeek open-sourced the reasoning models R1 and R1-Zero, which demonstrated capabilities similar to o1 across various domains at a fraction of the cost. Additionally, smaller distilled models were released, achieving high performance relative to their size.
- Google published a research paper on a new language model architecture called Titans, designed to enable models to retain both short- and long-term memory. This architecture significantly improves processing for extended context windows.
- DeepSeek open-sourced a fully multimodal model, Janus Pro 7B, which supports both text and image generation.
- Alibaba unveiled Qwen2.5-Max, a large language model that surpasses several leading models, including DeepSeek-V3, GPT-4o, and Claude 3.5. Additionally, the Qwen2.5-1M series was open-sourced, capable of processing up to one million tokens, along with the Qwen2.5-VL vision model series in three different sizes.