Google launched Gemini 3.7 Flash on Thursday, August 13, 2026, a new version of its fast, lower-cost AI model tuned for software coding and autonomous agent workflows. The release replaces Gemini 3.6 Flash, which Google shipped only three weeks earlier, and lands without the flagship Gemini 3.5 Pro that the company has kept listed as coming soon.

Google DeepMind positions the model for what it calls complex agentic tasks at scale. Reuters reported that the new model is designed for software coding and automated business tasks. Ars Technica reported that senior director Tulsee Doshi described the model as a workhorse built from core optimizations and developer feedback. 9to5Google added that the release is a direct result of developer feedback and an accelerated launch cadence.

Google lists four core capabilities for the model: agentic coding, advanced multimodal understanding across text, images, video, and audio, long-horizon task execution, and multi-step problem solving. The company has demonstrated Gemini 3.7 Flash orchestrating sub-agents to build interactive landing pages, powering a text-to-3D game tool called Antigravity, and training a robotics model through a multimodal graph loop.

Stronger Coding and Agent Benchmarks

Gemini 3.7 Flash posts sizable gains over its predecessor on the benchmarks Google publishes. Doshi told Ars Technica that the model's score on FrontierCode 1.1 Main, a production code quality test, rose from 34.4 percent to 43.6 percent. Its DeepSWE v1.1 long-horizon software engineering score climbed from 48.6 percent to 65.3 percent, and its WebDev Arena Elo rating rose from 1,538 to 1,588, the highest among the compared models.

The gains extend beyond coding. Google DeepMind reported that document comprehension on the GDP.pdf benchmark improved to 34 percent from 22 percent, and AutomationBench, which measures enterprise workflow automation, rose to 30.4 percent from 17 percent. On legal workflows the model reached 90.7 percent on Harvey LAB-AA, the strongest result in Google's comparison table.

The model also leads on several multimodal tests. On LVBench long video understanding it scored 85.4 percent, ahead of Claude Sonnet 5 at 68.5 percent and GPT-5.6 at 78.9 percent. On GDM-MRCR v2, a 128,000-token long-context benchmark, it reached 97.0 percent accuracy, the highest figure Google listed.

Seeking Alpha reported that Gemini 3.7 Flash beat Anthropic's Claude Sonnet 5 and OpenAI's GPT-5.6 on several coding evaluations. According to Artificial Analysis, the model sits at the Pareto frontier of intelligence versus speed, meaning it offers a strong balance of capability and latency for its price.

Pricing and the AI Price War

Google is pricing Gemini 3.7 Flash aggressively. The model carries an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, a rate that holds through December 31, 2026. Starting January 1, 2027, the price rises to $1.50 per million input tokens and $7.50 per million output tokens, according to Google DeepMind.

Neowin reported that the pricing places Google squarely in an escalating AI model price war, following recent rate cuts from OpenAI and competitive launches from rivals. Cryptopolitan described the model as priced at roughly half the cost of its predecessor for coding workloads.

Availability and Early Reactions

Gemini 3.7 Flash is available through Google AI Studio under the model identifier gemini-3.7-flash, in the Gemini app, and through the Gemini Spark assistant, according to Android Authority and 9to5Google. Enterprise customers described early results favorably. Box said the model was about 35 percent cheaper than 3.6 Flash in its testing, with a higher prompt-cache hit rate. Databricks, Emergent, and Nunu.ai also reported stronger performance on agentic, design, and document tasks.

The model competes directly with OpenAI's GPT-5.6 family and Muse's Spark 1.2. Decrypt reported that OpenAI's faster GPT-5.6 Sol Ultrafast remains invite-only, which gives Google a head start with a generally available low-cost model.

What Happens Next

Google has not announced a date for Gemini 3.5 Pro, which remains listed as coming soon. The company's accelerated cadence, two Flash releases in three weeks, suggests further updates are likely. Full benchmark methodology is published on the Google DeepMind site, and the model is live in Google AI Studio for developers to test.