[Digital Today reporter Chi-gyu Hwang] Google's AI research organisation will release a new model, Gemini 3.8 Flash, with sharply improved coding ability as soon as Sept. 2 local time.

According to a Wall Street Journal report, Gemini 3.8 Flash is called "Schemakey" internally. In comparative tests using Jetski, Google's in-house coding tool, engineers were reported to have preferred the model over Anthropic's Opus model.

If the model posts strong results on industry-standard benchmarks, it could answer questions surrounding Google DeepMind, the WSJ said.

Google DeepMind co-founder Demis Hassabis (데미스 하사비스) stepped down last month after a management reshuffle. His successor, Koray Kavukcuoglu (코레이 카부크추오글루), has been stressing to staff the need to move faster.

Google briefly took the lead in the race to develop frontier AI models when it released Gemini 3.0 in November last year. But it has been seen as falling behind OpenAI and Anthropic after showing multiple development setbacks this year.

As agentic coding has emerged as a core AI business area, Gemini has lagged behind Anthropic and OpenAI's flagship models.

Amid that, Google has since early this year put more research staff and computing resources into improving coding ability. It increased resources devoted to reinforcement learning, a post-training stage in which systems learn through trial and error. It recently hired Barrett Zoph (배럿 조프), who led OpenAI's post-training team, as a research vice president in charge of reinforcement learning and post-training.

Even if Flash 3.8 performs well, that alone does not make it easy to say Google has returned to the top-tier AI model race. The Flash series is designed to be small, cheap and fast, but it falls short of large models with trillions of parameters.

Google CEO Sundar Pichai (순다 피차이) said in May that the next model in the top-end "Pro" series would come out "next month", but the actual release has been delayed for months. The WSJ said 3.5 Pro candidates under internal review were scrapped because they were not sufficiently better than the Flash series.

The WSJ said the next flagship model, Gemini 4, posted good results in pre-training evaluations, but post-training work still remains.