Skip to content
NewsModel release

Google launches Gemini 3.7 Flash for coding and agents at half the price of 3.6 Flash

· by Pondero Newsdesk

The short version

Google released Gemini 3.7 Flash on August 13 at an introductory price of $0.75 per million input tokens, half of what 3.6 Flash cost at launch, with benchmark gains across coding, agentic, and knowledge tasks.

Google launches Gemini 3.7 Flash for coding and agents at half the price of 3.6 Flash

Gemini 3.7 Flash shipped August 13 at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, valid through December 31, 2026. That is half of what Gemini 3.6 Flash cost at its own launch three weeks earlier. Google also reported benchmark gains on every coding and agentic metric it publicly disclosed, per 9to5Google.

What

Google released 3.7 Flash across the Gemini API, Google AI Studio, Gemini Enterprise Agent Platform, Android Studio, and the consumer Gemini Spark product (requiring AI Pro or Ultra subscriptions). The model supports a 1 million token context window and up to 64,000 output tokens per prompt.

Benchmark numbers from the announcement showed consistent improvement over 3.6 Flash. On DeepSWE v1.1, a test of agentic code repair across real software repositories, 3.7 Flash scored 65.3 percent versus 49.0 percent for 3.6 Flash. On FrontierCode 1.1 Main, which covers 100 coding tasks requiring functional code, style compliance, and bug testing, the score moved from 34.4 to 43.6 percent. The WebDev Arena Elo score on Arena.ai reached 1,588, up from 1,538. Knowledge-dense tasks also moved: GDP.pdf processing (complex financial document comprehension) went from 22.0 to 34.0 percent, and AutomationBench (real-world business workflow completion) rose from 17.0 to 30.4 percent per 9to5Google.

Google claimed the model outperformed comparable models from Anthropic and OpenAI across nine benchmarks. On GDP.pdf specifically, 3.7 Flash finished 6 percent ahead of Claude Sonnet 5 and 9.3 percent ahead of GPT-5.6 Terra, per SiliconAngle. The full list of nine benchmarks was not published in the initial announcement.

Tulsee Doshi, Google's senior director of product management, described the model's behavioral improvements in the company's announcement blog post. The model "better adapts to roadblocks, clarifies intent when needed, and follows instructions with greater fidelity," and "thinks more diligently, putting in more effort into multi-step planning and tool calls," per SiliconAngle.

Why it matters

Google cut the input price by 50 percent relative to 3.6 Flash's launch rate while posting gains on the coding and agentic benchmarks most directly tied to production costs. DeepSWE v1.1 at 65.3 percent and AutomationBench at 30.4 percent measure exactly the task types that drive high token volumes in autonomous pipelines: issue resolution and multi-step workflow execution. For teams modeling costs against API usage at scale, $0.75 per million input tokens through December is a concrete budget input.

The introductory rate expires December 31. Google has not said what pricing looks like after that date. Teams committing large pipelines to 3.7 Flash should plan conservatively for a rate increase in Q1 2027. The nine-benchmark competitive claim is per Google; the full comparison set has not been independently confirmed, and third-party coverage of those specific evaluations is still developing.

What to watch next

Google has not disclosed the post-December 31 pricing for 3.7 Flash. If the $0.75/$3.75 rate holds or only rises modestly, it will put direct pressure on comparable Flash and Sonnet-class pricing from other labs. Independent third-party benchmark confirmation of the nine-eval claim, and the specific benchmarks beyond FrontierCode 1.1 Main and GDP.pdf, will clarify how 3.7 Flash stacks up outside Google's own evaluations.

Sources