Google Gemini 3.7 Flash targets coding and AI agents with 50% introductory price cut

Google has launched Gemini 3.7 Flash, the latest version of its high-volume artificial intelligence model, with a sharper focus on software development, AI agents and enterprise knowledge work. The release also comes with a temporary 50% reduction in API pricing, giving businesses several months to assess whether the model can lower the overall cost of running automated workflows.

The launch follows Gemini 3.6 Flash by just three weeks, highlighting the increasingly rapid development cycle behind Google’s Flash family of models.

Gemini 3.7 Flash arrives with lower introductory API pricing

For developers, Gemini 3.7 Flash is initially priced at $0.75 per million input tokens and $3.75 per million output tokens until 31 December 2026.

From 1 January 2027, those prices are due to double to $1.50 per million input tokens and $7.50 per million output tokens. Context caching will similarly rise from $0.075 to $0.15 per million tokens.

The introductory discount could be particularly significant for organisations operating coding assistants and autonomous business agents, where one task can require numerous model requests and tool interactions.

However, the headline token price is only part of the equation. For businesses, the more important measure will be the cost of completing a task successfully, including retries and any human intervention required.

Google focuses on coding and more reliable AI agents

Google describes Gemini 3.7 Flash as its “most intelligent workhorse model yet for coding and agents”.

More emphasis on planning and tool use

The company says the model is better at responding to obstacles, interpreting user intent and following instructions accurately. It is also designed to apply more effort to multi-stage planning and tool calls.

These improvements could be important for enterprise applications. An AI coding agent that can recover from errors and avoid unnecessary changes may require less developer supervision. Likewise, business agents working across documents and software systems need reliable tool selection to prevent workflows from failing.

Google DeepMind says Gemini 3.7 Flash has improved debugging and issue-resolution capabilities, while also generating more functional websites and applications with fewer prompts.

Coding benchmarks show substantial gains

Google’s benchmark results suggest significant improvements over Gemini 3.6 Flash, although the new model does not lead every test.

Gemini 3.7 Flash scored 43.6% on FrontierCode 1.1 Main, compared with 34.4% for Gemini 3.6 Flash. Google’s figures place Claude Sonnet 5 at 42.7% and GPT-5.6 Terra at 41.3%.

On DeepSWE v1.1, which evaluates longer software engineering tasks, Gemini 3.7 Flash achieved 65.3%, up from 49.0% for its predecessor. GPT-5.6 Terra remained ahead at 69.6%.

Web development also improved, with Gemini 3.7 Flash receiving an Elo score of 1588 on Code Arena. Gemini 3.6 Flash scored 1538, while Google lists Claude Sonnet 5 at 1541 and GPT-5.6 Terra at 1523.

The results are less decisive elsewhere. Gemini 3.7 Flash recorded 85.8% on Terminal-bench 2.1, behind GPT-5.6 Terra’s 87.4%. Claude Sonnet 5 also led Google’s Agent’s Last Exam multimodal desktop and operating-system comparison, recording a 33.3% pass rate against Gemini 3.7 Flash’s 26.3%.

Google’s figures therefore suggest a model that has become more competitive in coding and agent-based workloads rather than one that leads across every category.

Enterprise automation and document processing improve

The model’s improvements extend beyond software engineering.

On AutomationBench, which measures enterprise workflow automation, Gemini 3.7 Flash scored 30.4%, compared with 17.0% for Gemini 3.6 Flash. Google’s comparison lists Claude Sonnet 5 at 10.7% and GPT-5.6 Terra at 23.6%.

Gemini 3.7 Flash also achieved 34.0% on GDP.PDF, an evaluation involving complex PDF comprehension. Gemini 3.6 Flash scored 22.0%, Claude Sonnet 5 reached 28.0% and GPT-5.6 Terra recorded 24.7%.

These capabilities could matter for businesses using AI to process lengthy reports, extract information, interact with software tools and prepare documents for human review.

Google is incorporating the model into Gemini Spark for Google AI Pro and Ultra subscribers, with the company highlighting improved knowledge work and tool use across Google Workspace applications.

Enterprise customers can also access Gemini 3.7 Flash through Google’s Gemini Enterprise Agent Platform and Gemini Enterprise.

Competitive pricing adds pressure to the AI market

Gemini 3.6 Flash’s standard API price is already $1.50 per million input tokens and $7.50 per million output tokens.

Google’s comparison places Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens, while GPT-5.6 Terra is listed at $2 and $12 respectively.

For autonomous AI agents, however, cheaper tokens do not automatically mean cheaper operations. A lower-priced model requiring repeated attempts could ultimately cost more than a more expensive model completing tasks correctly the first time.

Google is effectively betting that improvements in execution reliability, combined with its temporary pricing discount, will make Gemini 3.7 Flash attractive for high-volume enterprise deployments.

Gemini 3.5 Pro delay remains a challenge for Google

The rapid progress of Google’s Flash models comes as questions continue over the company’s next flagship AI system.

Gemini 3.5 Pro has yet to receive a release date despite previously being described as undergoing partner testing. Google’s latest released general-purpose Pro model therefore remains Gemini 3.1 Pro, introduced in February.

Reports have suggested Gemini 3.5 Pro missed its initial timetable after falling short of internal targets, particularly in coding. Google is meanwhile developing Gemini 4, which is expected to become an important test of whether the company can regain broader leadership at the top end of the AI market.

The delays have coincided with significant changes to Google’s AI leadership. Google DeepMind co-founder Demis Hassabis has moved from day-to-day management of the division to become its chair and Alphabet’s chief scientist.

Former DeepMind chief technology officer Koray Kavukcuoglu now leads the unit as a senior vice-president reporting to Google chief executive Sundar Pichai, consolidating responsibility for Gemini model development, research, applications and developer operations.

Despite the disruption, Google retains significant advantages through products including Search, Workspace, Android and Google Cloud, as well as its custom AI infrastructure. The company says the Gemini app has exceeded 950 million monthly users.

Gemini 3.7 Flash available across Google’s developer platforms

Developers can access Gemini 3.7 Flash through the Gemini API, Google AI Studio, Android Studio and Google’s Antigravity environment. Enterprise deployment is available through Gemini Enterprise products, while eligible Google AI Pro and Ultra subscribers can use the model through Spark in supported markets.

Google is also introducing updated safeguards addressing chemical, biological, radiological and nuclear risks, alongside measures intended to reduce cyber-offence misuse.

The three-week gap between Gemini 3.6 Flash and Gemini 3.7 Flash demonstrates how quickly Google is now updating its efficiency-focused models. For businesses, however, benchmark improvements and discounted token prices will need to translate into reliable performance on their own codebases, documents and workflows.

Ultimately, Gemini 3.7 Flash’s competitive position will depend on how consistently it completes real-world tasks with minimal retries and human supervision — particularly once its introductory pricing ends in January 2027.