Meta and Google Release Top-Tier AI Models a Day Apart

Meta Superintelligence Labs unveiled Muse Wonder and the higher-tier Muse Wonder Full on October 12, while Google publicly released Gemini 4 Argon on October 13, according to tech42.co.kr. The back-to-back launches put two frontier models on the market within 24 hours.

Muse Wonder is available through the Meta model API and the Muse Code coding tool. Muse Wonder Full is provided only to Muse Code Power and Maximum subscribers and does not include API access.

On the Artificial Analysis intelligence index, Muse Wonder Full scored 56 and Muse Wonder scored 52, up from the 48 scored by Muse Spark 1.3, which launched about six weeks earlier. Gemini 4 Argon is available to developers through paid Gemini API plans and to general users through AI Pro and Ultra subscriptions in the Gemini app and Search AI mode.

The Gemini app added a feature that continues tasks when a conversation is converted into a 'task' even while the user is away. Linked apps now exceed 40, including Instacart and OpenTable.

Alibaba's Qwen team published the QwenGyre paper on October 12, reporting that the framework reduced the time to complete 48 reinforcement learning steps from 134.55 hours to 75.42 hours, about 44%, under the same GPU budget. The NL2RepoBench pass rate rose from 52.48% to 58.54% after training with QwenGyre.

Kakao announced Kanana Space, a KakaoTalk AI assistant summoned by swiping two fingers, and Canvas, a PC personal agent team builder, at the if(kakao)26 developer event on October 13-14. Both aim for closed beta testing within the year. Meanwhile, the government's MoA AI began closed beta testing on October 16 with SK Telecom, KT and Kakao consortia for participating companies and public institution officials, targeting a December official launch.