Thought Leadership
Google has launched Gemini 3.7 Flash just three weeks after its predecessor, with an introductory price cut of 50% through the end of 2026.
Google calls Gemini 3.7 Flash its “most intelligent workhorse model for coding and agents” to date, highlighting significant gains across several software engineering benchmarks. On FrontierCode 1.1 Main, which measures the quality of production ready code, Google reports that the model’s score rises from 34.4% to 43.6%. On the long horizon evaluation DeepSWE v1.1, performance increases from 49.0% to 65.3%. The model also delivers substantial improvements in web layout generation, according to Google, with its Elo rating in the WebDev Arena increasing from 1,538 to 1,588. In enterprise automation, measured using the AutomationBench benchmark, the score jumps from 17.0% to 30.4%.
Google also says the model is more flexible when dealing with obstacles in agent workflows, better at clarifying user intent when necessary and more reliable at following instructions than its predecessor. The company has also further strengthened safeguards against misuse involving chemical, biological, radiological and nuclear risks, as well as cyberattacks, under its Frontier Safety Framework.
Results Mixed Against Competing Models
A direct comparison with models from other providers shows no consistent lead for Gemini 3.7 Flash in Google’s own published benchmark tables. On Terminal-Bench 2.1, Google’s model scores 85.8%, below OpenAI’s GPT-5.6 Terra at 87.4%. According to the published data, GPT-5.6 Terra also leads on Terminal-Bench 3.0 and OSWorld-2.0.
In multimodal desktop and operating system tasks in the Agent’s Last Exam benchmark, however, Anthropic’s Claude Sonnet 5 comes out ahead with a success rate of 33.3%, compared with 26.3% for Gemini 3.7 Flash. The results therefore point less to a model that is superior across the board and more to a significantly more competitive offering at a lower price point.
Introductory Pricing Ends at Year-End
Through December 31, 2026, Gemini 3.7 Flash costs $0.75 per million input tokens and $3.75 per million output tokens, both half the original prices of Gemini 3.6 Flash. Starting January 1, 2027, the rates will double to $1.50 and $7.50, respectively.
For development teams deploying the model at scale for coding or business agents, that leaves a limited window to determine whether the improvements Google reports in error correction and reduced manual oversight actually translate into lower total operating costs.
The new model is available through Google AI Studio, Android Studio, the Gemini Enterprise Agent Platform, the Gemini Enterprise app and Google Antigravity in more than 160 countries. In the consumer Gemini app, the model also powers Spark, the always on agent feature, which requires an AI Pro or Ultra subscription.
The launch comes as Google’s next flagship model, Gemini 3.5 Pro, still has no specific release date, despite CEO Sundar Pichai announcing in May that it would arrive the following month. Google has meanwhile confirmed that it is already working on the successor, Gemini 4.
(Editorial Team)
