Google launches Gemini 3.7 Flash with coding gains, half price

APPSO · wechat · 2026-08-14

Google has released Gemini 3.7 Flash, a new model focused on coding and agent workflows, priced at half the original cost of 3.6 Flash. It scores 43.6% on FrontierCode 1.1 Main, surpassing Claude Sonnet 5 (42.7%) and GPT-5.6 Terra (41.3%), and achieves 1588 Elo on WebDev CodeArena, also ahead. On DeepSWE v1.1 it reaches 65.3%, still below GPT-5.6 Terra's 69.6%. In enterprise automation (AutomationBench), it jumps from 17.0% to 30.4%, beating Claude Sonnet 5 (10.7%) and GPT-5.6 Terra (23.6%).

The model improves in multi-step planning, tool calling, and instruction following, and supports multimodal applications like generating 3D games, web development, and robot training. Google also announced that Gemini Spark will switch to 3.7 Flash, enhancing Workspace tool use.

The release follows a leadership change at Google DeepMind, with Koray Kavukcuoglu taking over. Reuters reports co-founder Sergey Brin has been pressuring the team to accelerate Gemini development and push recursive self-improvement. Google faces delays in flagship models and compute constraints, so Flash series aims to compete with frequent updates and low pricing.

Related event: Google Launches Gemini 3.7 Flash with Enhanced Coding and Agent Capabilities at Half Price(43 posts)→

Original post →

More from Companies & People

Companies & People channel →