Google launches Gemini 3.7 Flash for coding and agent workflows

Google has introduced Gemini 3.7 Flash, a new model aimed at software engineering, web development, knowledge work, and AI agents. The release arrives only three weeks after Gemini 3.6 Flash and is positioned as a more capable “workhorse” for multi-step tasks.

The company reports improvements in debugging, issue resolution, first-pass code accuracy, and production-oriented software generation. On the benchmarks cited by Google, Gemini 3.7 Flash scored 43.6% on FrontierCode 1.1 Main versus 34.4% for 3.6 Flash, and 65.3% on DeepSWE v1.1 versus 49.0%.

For builders creating interfaces and applications, Google says the model can produce more complete web apps in fewer prompts and better match reference screenshots, images, and design systems. Its WebDev Arena score rose to 1588, compared with 1538 for the previous model.

The upgrade also targets document-heavy and business workflows. Google reports higher results on GDP.pdf, which tests complex document comprehension, and AutomationBench, a benchmark for real-world workflow completion. The model is also designed to handle roadblocks, ask for clarification when needed, follow instructions more closely, and plan tool calls with greater care.

Lower introductory pricing

Gemini 3.7 Flash is available at an introductory rate through the end of 2026: $0.75 per million input tokens and $3.75 per million output tokens. Google says the input and output rates are half the original Gemini 3.6 Flash cost per million tokens.

For AI developers, the combination of stronger coding performance, fewer expected retries, multimodal UI generation, and lower initial pricing could make the model attractive for agent loops and production prototypes. As always, the benchmark gains are Google’s reported results, so teams should test performance against their own workloads.

Source: Google DeepMind Blog