Google's Gemini 4 Nears Perfect ARC-AGI Score as Open-Source Camp Debuts New 501B Model · 1017-741

2026-10-17 00:00

Author: RadarAI Editorial Editor: RadarAI Editorial Last updated: 2026-10-17 Review status: Editorial review pending Brief 速报官方 AI动态开源

Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.

## 🔍 Key Insights This week the AI industry saw a **dual explosion** in **model capabilities** and **infrastructure**: **Google Gemini 4 Argon** became the second lab to cross the 95% threshold on ARC-AGI-3 with a score of **97.5%** [5], while **Reflection** released its first open-weight model **Beam**, shaking up the open-source ecosystem with a **501B-parameter MoE** architecture under the Apache 2.0 license [14][23]. Meanwhile, real-world laptop testing of **Nvidia's RTX Spark** chip and **Doubao Work**'s one-click AI asset migration feature signal that on-device AI and user data portability are becoming new competitive focal points [0][10][11]. ## 🚀 Top Stories - **Gemini 4 Argon hits 97.5% on ARC-AGI-3** [5]: Google DeepMind becomes the second lab to cross 95%, at a cost of $4,800, but still trails GPT-6 Astra. - **Reflection releases 501B-parameter MoE model Beam** [14][23]: Its first open-weight model, Apache 2.0 licensed, 23B active parameters, targeting coding and long-horizon agent tasks. - **Nvidia RTX Spark laptop first hands-on test** [0][10]: Five laptops including the ASUS ProArt 16 can run a 27-billion-parameter large model locally; The Verge conducted a week-long in-depth test. - **Doubao Work launches one-click AI asset migration** [11]: Supports importing conversations, skills, and memory from ChatGPT and Claude Code; tested to complete in one to two minutes with context continuity. - **MAI-Transcribe-2 tops OpenRouter transcription leaderboard** [1][2]: Microsoft's MAI becomes the most accurate and most popular transcription model on OpenRouter, now available to try. - **Perplexity open-sources SPACE sandbox environment** [21]: Built on Firecracker microVMs, providing isolated security and long-lifecycle execution environments for AI Agents. - **Kling 4.0 fully rolled out** [9]: Kling AI has fully launched Kling 4.0 on kling.ai, offering finer motion control and professional-grade audio-video output. - **Meta's new commercial terms said to allow free scraping of publisher content** [19]: Effective October 30, legal experts believe it may bypass blocking directives such as robots.txt; Meta denies any change in its crawling methods. ## 🔗 Sources [0] Nvidia RTX Spark laptop first hands-on test, I achieved WorkBuddy quota freedom — https://www.bestblogs.dev/article/8fca76c0ba?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [1] MAI-Transcribe-2 launches on OpenRouter — https://aihot.news/items/q5zc74jblm1ayo630xtpe34i9 [2] MAI tops OpenRouter transcription model leaderboard — https://aihot.news/items/e79z4iorx69fa9w4a9drfs4y0 [5] Google Gemini 4 Argon hits 97.5% on ARC-AGI-3, but still trails the leaders in math and software engineering — https://aihot.news/items/cqlsedakjd026b9gwsdor3la5 [9] Kling 4.0 fully rolled out on kling.ai; Kling AI discusses AI video deployment with WPP and Adobe at Advertising Week New York 2026 — https

This week the AI industry saw a dual explosion in model capabilities and infrastructure: Google Gemini 4 Argon became the second lab to cross the 95% threshold on ARC-AGI-3 with a score of 97.5% [5], while Reflection released its first open-weight model Beam, shaking up the open-source ecosystem with a 501B-parameter MoE architecture under the Apache 2.0 license [14][23]. Meanwhile, real-world laptop testing of Nvidia's RTX Spark chip and Doubao Work's one-click AI asset migration feature signal that on-device AI and user data portability are becoming new competitive focal points [0][10][11].

🚀 Top Stories

Gemini 4 Argon hits 97.5% on ARC-AGI-3 [5]: Google DeepMind becomes the second lab to cross 95%, at a cost of $4,800, but still trails GPT-6 Astra.
Reflection releases 501B-parameter MoE model Beam [14][23]: Its first open-weight model, Apache 2.0 licensed, 23B active parameters, targeting coding and long-horizon agent tasks.
Nvidia RTX Spark laptop first hands-on test [0][10]: Five laptops including the ASUS ProArt 16 can run a 27-billion-parameter large model locally; The Verge conducted a week-long in-depth test.
Doubao Work launches one-click AI asset migration [11]: Supports importing conversations, skills, and memory from ChatGPT and Claude Code; tested to complete in one to two minutes with context continuity.
MAI-Transcribe-2 tops OpenRouter transcription leaderboard [1][2]: Microsoft's MAI becomes the most accurate and most popular transcription model on OpenRouter, now available to try.
Perplexity open-sources SPACE sandbox environment [21]: Built on Firecracker microVMs, providing isolated security and long-lifecycle execution environments for AI Agents.
Kling 4.0 fully rolled out [9]: Kling AI has fully launched Kling 4.0 on kling.ai, offering finer motion control and professional-grade audio-video output.
Meta's new commercial terms said to allow free scraping of publisher content [19]: Effective October 30, legal experts believe it may bypass blocking directives such as robots.txt; Meta denies any change in its crawling methods.

🔗 Sources

[0] Nvidia RTX Spark laptop first hands-on test, I achieved WorkBuddy quota freedom — https://www.bestblogs.dev/article/8fca76c0ba?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [1] MAI-Transcribe-2 launches on OpenRouter — https://aihot.news/items/q5zc74jblm1ayo630xtpe34i9 [2] MAI tops OpenRouter transcription model leaderboard — https://aihot.news/items/e79z4iorx69fa9w4a9drfs4y0 [5] Google Gemini 4 Argon hits 97.5% on ARC-AGI-3, but still trails the leaders in math and software engineering — https://aihot.news/items/cqlsedakjd026b9gwsdor3la5 [9] Kling 4.0 fully rolled out on kling.ai; Kling AI discusses AI video deployment with WPP and Adobe at Advertising Week New York 2026 — https

← Back to Updates