Google launches Gemini 3.7 Flash for coding, AI agent projects
SiliconANGLE Maria Deutscher ● Covered by 7 sources
Google just shipped Gemini 3.7 Flash, its strongest cheap model yet, for coding and AI agents. It says it beats rival models on nine tests and costs half as much as 3.6 Flash through year-end.
Based on reporting by SiliconANGLE, Maria Deutscher — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google has launched Gemini 3.7 Flash, a new entry-level model aimed at coding and AI agent work. The company says it arrived just three weeks after Gemini 3.6 Flash, but still brings a noticeable jump in output quality.
Google says the model beat comparable offerings from Anthropic and OpenAI across nine benchmarks. One of the standout results came on FrontierCode 1.1 Main, a test of 100 programming tasks across multiple languages. That benchmark is stricter than a simple code dump: the model has to write working code, test it for bugs, and follow project-specific style rules. Google says Gemini 3.7 Flash did especially well at generating user interfaces, with layouts that matched user-uploaded reference images more closely than before.
The model can handle up to 1 million tokens of images, video and text in a single prompt, and it can return as many as 64,000 tokens of text. Tulsee Doshi, Google’s senior director of product management, said in a blog post that the model adapts better when it hits roadblocks, asks for clarification when needed, and follows instructions more faithfully. She also said it spends more effort on multi-step planning and tool calls.
Google is pitching the model as more than a coding assistant. On its GDP.pdf benchmark, which checks how well systems answer questions about business documents, Gemini 3.7 Flash scored 34% and came out 6% ahead of Claude Sonnet 5 and 9.3% ahead of GPT-5.6 Terra. Google also says the model can power ensembles of AI agents.
The company did not spell out the architecture or training method. The model card says it uses the same architecture as Gemini 3.6 Flash, which itself comes from Gemini 3 Pro and its transformer-based mixture-of-experts design. Developers can access Gemini 3.7 Flash at half the price of Gemini 3.6 Flash through the end of the year, and Google is also bringing it into Gemini Spark, its consumer AI agent that launched in March and can browse the web and act inside other Google services.
My take — AI-written commentary, not fact-checked reporting
Google is doing the obvious thing here: undercut the market on price while trying to make “cheap” sound like “best.” That’s not cynical; it’s the playbook. The real tell is the push into agents and UI work, where polished demos still matter almost as much as raw scores.
Read more about this at: SiliconANGLE
