Google and Meta Simultaneously Launch AI Models Amid Competitive Race

Google and Meta unveiled their latest frontier AI models just hours apart on Wednesday. Google introduced Gemini 3.8 Flash alongside a cybersecurity variant, while Meta launched Muse Spark 1.3.
The simultaneous releases have sparked direct comparisons. Independent test results from Artifical Analysis show mixed outcomes: Meta leads in agent-based knowledge tasks and scientific reasoning, while Google excels in fact recall and terminal coding.
Gemini 3.8 Flash is Google's third flash model release in six weeks. Pricing stands at $0.75 per million input tokens and $3.75 per million output tokens, valid until December 31, 2026. After that date, prices will double to $1.50 and $7.50 respectively.
Google also launched Gemini 3.8 Flash Cyber, a cybersecurity-focused model designed for trusted defense groups. It scored 86.2% on the CyberGym security vulnerability benchmark and 47.2% on the CWE-bench patch benchmark, with Google claiming it provides 2.6x more correct patches for Chrome vulnerabilities than commercial large models.
Access to Gemini 3.8 Flash Cyber is restricted through the Fairwind Program, limited to government agencies and major infrastructure operators. OpenAI previously applied similar restrictions to its Astra model after it reached a critical cybersecurity threshold.
Meanwhile, Meta released Muse Spark 1.3 via Muse Code and Meta Model API. Engineers reported approximately a 20% reduction in tool calls compared to version 1.2.
Independent benchmark results show mixed performance: Muse Spark 1.3 scored 1,754 Elo on the GDPval-AA v2 scale (maximum mode), while Gemini 3.8 Flash (premium version) scored 1,545. Meta led in Sierra Research Bank Agent Test (52.4% vs. 44.9%) and CritPt physical reasoning tests. Gemini excelled in Terminal-Bench 2.1 (87.6%), AA-LCR long context (81%), and AA-omniscience accuracy (55%).
Gemini 3.8 Flash achieved the highest GPQA Diamond score at 95% among tested models, while both models scored nearly identically on Humanity's Last Exam with a one-point difference.
Meta's top-scoring model is currently unavailable as additional safety verification completes; they are offering the xhigh version instead. This version scored 61 points on the Artificial Analysis Intelligence Index, surpassing Muse Spark 1.2 by four points but trailing Claude Haiku 5.1 (66) and Claude Opus 5 (63).
More model releases are planned: Elon Musk announced Grok 4.7 will launch soon, with four major frontier models expected within two weeks.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 구글·메타 경쟁 AI 모델 동시 공개…승자는?