Google Releases Another Lightweight AI Model After Three Weeks, but Trails Meta in Evaluation

Google has released another lightweight artificial intelligence model just three weeks after its last one, intensifying the race to update fast. Yet an independent evaluation reported that the new Google model scored below Meta’s latest offering.

Google unveiled Gemini 3.8 Flash on the second day of the month, presenting it as a model with improved reasoning and coding capabilities. The company had introduced Gemini 3.6 Flash on July 21 and Gemini 3.7 Flash on the 13th of the previous month. That means Google has released three lightweight models in six weeks. Gemini 3.8 Flash will be integrated into Google AI Studio and the Gemini application.

Google emphasized that the model exceeded some larger systems in coding tests despite its smaller size. On the DeepSWE 1.1 benchmark, which measures the ability to solve software-engineering problems over extended periods, Google said Gemini 3.8 Flash scored higher than OpenAI’s top model, GPT-5.6 Sol, and Anthropic’s Claude Opus 5. The system can increase its reasoning steps on complex tasks to improve accuracy.

The outside assessment was less favorable. AI model analysis company Artificial Analysis gave Meta’s Muse Spark 1.3, released the same day, an intelligence score of 62. That matched the high-performance setting of Anthropic’s Claude Fable 5.1. Gemini 3.8 Flash received a top score of 59, three points lower. Meta described Muse Spark 1.3 as an improved model for coding and agent tasks that use multiple tools. Compared with its predecessor, it reduced tool calls by about 20% and token use by about 25%.

While Google has accelerated its lightweight releases, its high-performance lineup has slowed. Gemini 3.5 Pro had been expected in June but remains in partner testing. The last new model in Google’s Pro series was Gemini 3.1 Pro, released in February.