Google’s Gemini 3.7 Flash targets coding agents

Google has released Gemini 3.7 Flash, roughly three weeks after Gemini 3.6 Flash, with a focus on coding, agent workflows, web development, and document-heavy business tasks. Android Authority reports that the model is already replacing 3.6 Flash in Gemini Spark, while VentureBeat says Google is pairing the upgrade with temporarily reduced API pricing.

“Our most intelligent workhorse model yet for coding and agents.”

Google says developer feedback and algorithmic improvements helped Gemini 3.7 Flash handle debugging and issue resolution more accurately, produce more functional web layouts, and generate feature-complete applications in fewer prompts. It also claims better performance on finance, law, biosciences, PDF comprehension, and enterprise knowledge work.

The model is designed to recover more effectively when an agent encounters a roadblock, clarify intent when necessary, follow instructions more closely, and apply more effort to multi-step planning and tool calls. In practice, that could mean fewer retries and less human intervention, although the company’s claims still need to be tested against production workloads.

Recommended reading

DeepSeek launches Harness as API prices jump

Gemini 3.7 Flash benchmark results

Google’s published results show substantial gains over Gemini 3.6 Flash in several coding and enterprise evaluations, but not universal leadership:

• FrontierCode 1.1 Main: 43.6%, up from 34.4% for Gemini 3.6 Flash. Google lists Claude Sonnet 5 at 42.7% and GPT-5.6 Terra at 41.3%.
• DeepSWE v1.1: 65.3%, compared with 49.0% for 3.6 Flash. GPT-5.6 Terra remains ahead at 69.6%.
• Code Arena: an Elo score of 1588, ahead of 3.6 Flash at 1538, Claude Sonnet 5 at 1541, and GPT-5.6 Terra at 1523.
• AutomationBench: 30.4%, versus 17.0% for 3.6 Flash, 10.7% for Claude Sonnet 5, and 23.6% for GPT-5.6 Terra.
• GDP.PDF: 34.0%, compared with 22.0% for 3.6 Flash, 28.0% for Claude Sonnet 5, and 24.7% for GPT-5.6 Terra.

The results are weaker on some broader agent evaluations. Gemini 3.7 Flash scores 85.8% on Terminal-bench 2.1, below GPT-5.6 Terra’s 87.4%. On Agent’s Last Exam, Claude Sonnet 5 leads with a 33.3% pass rate, compared with 26.3% for Gemini 3.7 Flash.

That makes the release more targeted than a blanket claim of model leadership: Google is showing stronger performance in coding, web development, PDF comprehension, and workflow automation, while competitors still lead some general agent tasks.

Temporary API discount through 2026

Gemini 3.7 Flash costs $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Context caching costs $0.075 per million tokens during the introductory period.

On January 1, 2027, those prices double to $1.50 per million input tokens, $7.50 per million output tokens, and $0.15 per million cached tokens. The future standard rate matches the pricing Android Authority reports for Gemini 3.6 Flash, so the discount is temporary rather than a permanent reduction in Google’s Flash pricing.

VentureBeat’s comparisons place 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 agent deployments, however, token price alone will not determine operating cost: a cheaper model that needs more retries can cost more per completed task.

Availability and Google’s wider model strategy

Developers can access Gemini 3.7 Flash through the Gemini API in Google AI Studio, Android Studio, and Google Antigravity. Enterprise customers can use it through the Gemini Enterprise Agent Platform and Gemini Enterprise, while Google AI Pro and Ultra subscribers can access it through Spark in supported countries.

Google says Spark will use the model for knowledge work and tool-based workflows across Google Workspace, including consolidating files, drafting emails, and updating status documents. The company is also shipping updated safeguards for chemical, biological, radiological, and nuclear risks, as well as cyber-offense misuse.

The rapid release comes while Google’s next flagship Pro model remains unresolved. VentureBeat reports that Google provided no release date for Gemini 3.5 Pro, which had been described as undergoing partner testing. The outlet also notes that Google’s latest broadly released general-purpose Pro model remains Gemini 3.1 Pro, introduced in February.

For now, Gemini 3.7 Flash gives developers a lower-cost model with strong results on several practical workloads. The key commercial test will arrive after January 1, 2027, when teams must determine whether its improved first-pass accuracy reduces total task costs enough to justify the return to standard pricing.