Gemini 3.7 Flash Arrives as Google’s Frontier Model Delay Continues

Google has announced Gemini 3.7 Flash, a new version of its Flash model family with improvements in coding and agentic workflows. 

The model arrives just three weeks after Gemini 3.6 Flash, which was released on July 21. 

It is also being offered at introductory prices of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens, half the standard rates for Gemini 3.6 Flash—$1.50 and $7.50, respectively. 

The introductory pricing is valid through December 31, after which the standard rates apply.

The model is also optimised for speed and, according to Artificial Analysis, is one of the fastest AI models, delivering 340 tokens per second via the first-party API. 

Besides, Arena AI, which evaluates AI models on coding tasks using both performance and pricing data, places Gemini 3.7 Flash on the Pareto frontier of Code Arena. 

In this context, the frontier represents models that offer a competitive balance between coding performance and cost, with no other model providing both a higher score and a lower price.

Anastasios Angelopoulos, Co-Founder and CEO of Arena AI, said on X, “This makes Google DeepMind the only US lab other than Anthropic that’s on the Pareto frontier of Code Arena.” 

The launch comes as the arrival of Google’s next flagship model, Gemini 3.5 Pro, remains delayed. A report from SemiAnalysis said Google had put the model on hold and was instead looking towards Gemini 4. 

Google has already confirmed that Gemini 4 is in pre-training. 

The launch also comes amid a major shake-up at Google DeepMind. Jeff Dean, Google’s longtime chief scientist and a co-leader of Gemini’s development, left the company after 27 years alongside several senior AI researchers to start a new AI venture. 

Demis Hassabis has also stepped down as DeepMind CEO to become chairman and Alphabet’s Chief Scientist, while Koray Kavukcuoglu has taken over day-to-day leadership of the lab.