Google today launched Gemini 3.7 Flash, an entry-level foundation model tuned for coding and multi-step AI agent workflows, arriving just three weeks after Gemini 3.6 Flash and undercutting its predecessor by 50% on price. The model rolls out immediately through the Gemini API, AI Studio, Gemini Enterprise, and the consumer-facing Spark agent for Pro and Ultra subscribers.
Coding Benchmarks Google Says Beat Anthropic And OpenAI
Gemini 3.7 Flash jumps from 34.4% to 43.6% on FrontierCode 1.1 Main, a 100-task benchmark covering production coding requirements, and from 49% to 65.3% on the DeepSWE v1.1 long-horizon software-engineering test. On GDP.pdf, a business-document QA benchmark, it scores 34%, ahead of Claude Sonnet 5 by six points and GPT-5.6 Terra by 9.3 points. The model retains the 1M-token multimodal context window of Gemini 3.6 Flash and returns up to 64,000 output tokens per prompt.
50% Price Cut Signals A Race For Agent Workloads
Launch pricing sits at $0.75 per million input tokens and $3.75 per million output tokens through year-end, half the previous Flash cost. Google is positioning the discount at developers building multi-step agents, where token consumption compounds quickly. The move lands the same week Google's Gemini assistant crossed one billion monthly users and comes as Anthropic builds in-house chip teams and DeepSeek unveils V4 Flash to compete on inference economics.
Consumer Agent Push With Spark
Google is also porting Gemini 3.7 Flash to Spark, the agent that debuted in March 2026 and can browse the web and act inside Google services. Tulsee Doshi, senior director of product management, said the model “better adapts to roadblocks, clarifies intent when needed, and follows instructions with greater fidelity,” a direct pitch at reliability metrics that matter for enterprise agent deployments. See related coverage in our xAI Grok 4.6 launch report.
Reporting based on coverage from SiliconANGLE, Reuters and Google's DeepMind model card.