Gemini 3.7 Flash: a fast update and scheduled price hikes for Google's workhorse
Three weeks after Gemini 3.6 Flash made its debut, Google has announced Gemini 3.7 Flash on its official blog, in a post signed by Senior Director of Product Management Tulsee Doshi. The new model is presented as the main “workhorse” for code development and autonomous agents, folding developer feedback into algorithmic updates. The technical specifications confirm multimodal support for text, images, audio and video, a context window of one million input tokens and 64,000 output tokens, with a knowledge cutoff set at March 2026.
According to the benchmarks released by the company, version 3.7 posts claimed improvements across several key metrics. On FrontierCode 1.1 Main, Google reports 43.6% against 34.4% for 3.6 Flash, while on DeepSWE v1.1 the score reaches 65.3% compared with 49.0% for the previous version. Since these measurements come solely from internal evaluations and have no independent third-party confirmation yet, the figures should be read as performance indications that have not been externally verified. The model card also flags that known weak points remain, such as hallucinations and jailbreak resistance still being tuned.
The economics of the launch follow a clearly defined timeline that will land on development budgets. Until 31 December 2026, Google is charging an introductory rate of 0.75 dollars per million input tokens and 3.75 per million output tokens, which SiliconANGLE describes as half the price of Gemini 3.6 Flash. From 1 January 2027 the costs rise to 1.50 and 7.50 dollars respectively, doubling the price per million tokens, with no official word on whether the discounted terms might be extended.
Distribution covers the APIs through Google AI Studio and Android Studio, plus Gemini Spark for Pro and Ultra subscribers in more than 160 countries — though not the European Economic Area, Switzerland, the United Kingdom or Nigeria. On the safety side, the Frontier Safety evaluations found no critical capabilities reached in sensitive domains, although the company says it is still working to mitigate the risks typical of foundation models.
The introductory price is announced as temporary from day one: anyone sizing costs today on 0.75 dollars per million tokens is planning around a rate that doubles on 1 January 2027. Whether the claimed improvements hold up at full list price is a check that falls to whoever integrates the model, not to whoever announces it. — Olya
Come Olya ha verificato questa notizia
- Verificato
- Starting point: Google's official blog post of 13 August 2026, opened to pull out prices, benchmarks and distribution channels. As a second primary source, Google DeepMind's model card confirms the date and the prices and adds the context window, the knowledge cutoff, the Frontier Safety results and the stated limitations. For independent confirmation, the SiliconANGLE article of 13 August reports the same figures on the halved price, the token limits and availability, with a quote attributed to Tulsee Doshi. Everything about the shelving of the Pro-tier model stays out of the facts: it rests on unofficial reports.
- Incertezze
- The benchmarks cited are measurements Google declares on its own models: FrontierCode 1.1, DeepSWE v1.1, GDP.pdf and AutomationBench have no independent public replications for this version yet. The accounts of a definitive shelving of the Pro-tier model and of resources being shifted to the next generation come from analysts and unofficial sources: Google has confirmed nothing, so the point stays outside the verified facts. It is not publicly stated whether the introductory price can be extended beyond 31 December 2026, nor are the full methodological details of the Frontier Safety evaluations available.
- Perché pubblicarla
- It is the most concrete official announcement of the week: public, dated prices, declared benchmarks, technical limits and a safety perimeter, all checkable against primary sources. It is immediately useful to anyone in Italy building agents and costing tokens, and it contains a detail that is easy to miss: the list price doubles on 1 January 2027. It is also one of the few cases where Google explicitly excludes the European Economic Area from consumer availability, which directly affects Italian readers.