In an unexpected move, just three weeks after the introduction of version 3.6, Google unveiled its latest AI model named Gemini 3.7 Flash. Although many users were waiting for the introduction of version 3.5 Pro, Google has for now focused on optimizing its small and fast models. This new model, which is the result of direct feedback from developers and core optimizations, shows a significant performance leap in areas such as coding and performing autonomous (agentic) tasks.
According to Tulsee Doshi, a senior director at Google, this model has shown stunning growth in various coding benchmarks, including FrontierCode and DeepSWE, with its score in the DeepSWE test rising from 49 to over 65 percent. In addition to coding power, Gemini 3.7 Flash has also improved in processing complex documents, rising from 17 to over 30 percent in workflow automation benchmarks. Google intends to consolidate its position in the market of cheap and efficient models by offering competitive initial pricing.