이미지: X — 뉴스 앰프 화면 갈무리
Summary
- Google has unveiled Gemini 3.7 Flash, a model tailored for coding and agentic tasks
- It scored 43.6% on the FrontierCode 1.1 benchmark, ahead of Claude Sonnet 5 (42.7%) and GPT-5.6 Terra (41.3%)
- Priced at $0.75 per million input tokens, it will be offered at half the price of 3.6 Flash through the end of the year
- 모델명
- Gemini 3.7 Flash
- 개발사
- Google Gemini
- FrontierCode 1.1 점수
- 43.6% (3.6 Flash 34.4%, Claude Sonnet 5 42.7%, GPT-5.6 Terra 41.3%)
- 입력 가격
- 100만 토큰당 0.75달러
- 가격 정책
- 연말까지 3.6 Flash 대비 50% 인하
What was announced
Google has rolled out Gemini 3.7 Flash, a new model aimed at coding and agentic tasks. According to a chart released alongside Google DeepMind's evaluation methodology document, the model scored 43.6% on the FrontierCode 1.1 benchmark, which measures real production-level code quality. Under the same conditions, Anthropic's Claude Sonnet 5 scored 42.7% and OpenAI's GPT-5.6 Terra scored 41.3%. Its predecessor, Gemini 3.6 Flash, scored 34.4% — meaning the new generation jumped roughly 9 percentage points. Pricing was set at $0.75 per million input tokens, and Google said the model will be available at half the price of 3.6 Flash through the end of the year. The company also noted improvements in debugging accuracy, agentic workflow stability, web development output, and reasoning accuracy.
What this means
"Flash" refers to the practical tier within Google's Gemini lineup, designed to respond faster and more lightly than the pro-grade models. It trades a bit of accuracy for lower cost and speed, making it suited for high-frequency use cases like code autocompletion or repetitive agentic tasks. What stands out about this release is how quickly it appeared. Just a day earlier, on August 13, one account reported spotting the model name "gemini-3.7-flash" exposed in a Google service filter search box. A day before that, on August 12, a banner advertising a "3.6 Flash" trial appeared on the Gemini Enterprise screen. In other words, signs of a transition from 3.6 to 3.7 were observed in quick succession over three days before culminating in the actual launch — a signal that Google appears to be running notably short revision cycles for its Flash lineup. It's also notable that the model narrowly outperformed rival models on a coding benchmark. Since metrics measuring production code quality tend to reflect the complex demands of real-world development more than simple problem-solving benchmarks, this could translate into a tangible difference for developers using agentic coding tools.
Why it matters
The practical takeaway from this release is that developers can now get a faster, cheaper coding model for the same money. In particular, since the half-price policy runs through the end of the year, this creates an opportunity to cut costs for those running agentic pipelines or automation tools that rely on frequent repeated calls. That said, details such as output token pricing or the timing of full API rollout were not clearly confirmed in this announcement. Given how quickly Google has been cycling through its Flash lineup, the next version update may not be far off either.
