Google AI Just Released Gemini 3.7 Flash: A Coding and Agent Model at $0.75/1M Input Tokens

August 13, 2026 | Source: MarkTechPost

Tags: Gemini, Google DeepMind, Flash models, coding AI, AI agents, LLM pricing, benchmarks

MarkTechPost's technical breakdown of Gemini 3.7 Flash highlights $0.75/1M input pricing, a 1M context window, and coding gains (FrontierCode 43.6%, DeepSWE 65.3%, AutomationBench 30.4%) while noting GPT-5.6 Terra still leads on DeepSWE at 69.6%.

Details

MarkTechPost provides the most technically detailed breakdown of the Gemini 3.7 Flash release. The model is a refinement of 3.6 Flash via algorithmic improvements — not a new pretraining run — preserving the same parameter count, 1M-token context window (64K output), and March 2026 knowledge cutoff while cutting pricing to $0.75/1M input and $3.75/1M output tokens. Key benchmark gains over 3.6 Flash: FrontierCode 1.1 Main (34.4% to 43.6%), DeepSWE v1.1 (49.0% to 65.3%), WebDev Arena Elo (1538 to 1588), GDP.pdf document comprehension (22.0% to 34.0%), AutomationBench business workflows (17.0% to 30.4%). Long-context retrieval on GDM-MRCR v2 at 128k reaches 97.0%. The article provides important competitive counterbalance: GPT-5.6 Terra leads on DeepSWE (69.6% vs 65.3%), Terminal-bench 2.1 (87.4%), and OSWorld-2.0 (50.2%). On knowledge work (GDPval-AA v2), Claude Sonnet 5 scores 1598 and Muse Spark 1.2 scores 1628 versus Gemini 3.7 Flash at 1525. The model shows a minor regression on CharXiv Reasoning (84.5% vs 85.2% for 3.6 Flash). Availability: Gemini API, Google AI Studio, Antigravity, Android Studio, Gemini Enterprise Agent Platform, and Gemini Enterprise app. Consumer access via Gemini Spark on AI Pro and Ultra plans. No self-hosting option. Customizable thinking configurations allow quality-cost-latency trade-offs per task.