Google has launched the Gemini Decisions API, powered by Gemini Decision 1. This new model returns calibrated probabilities for up to 1,000 yes/no, choice, and rubric questions across diverse inputs. The specialized API costs $0.10 per 1 million input tokens, adding to Google’s quickly growing Gemini model family, as reported by MarkTechPost.

Gemini Decision 1 follows a rapid succession of Google model releases and API expansions. On October 1, the company unveiled Gemini 4 Argon. It initially restricted access over safety concerns, then made its API available to developers by October 14. On October 29, Google introduced Gemini 4 Carbon, a model for cybersecurity and coding tasks. Gemini 4 Carbon has also now opened to developers and enterprises. It handles complex tasks like software development and financial research, according to The Economic Times.

This quick cadence of specialized Gemini models, each with distinct capabilities and an API-first rollout, shows Google’s aggressive push to commercialize its AI research. Our prior coverage on October 9 noted Google had begun charging for advanced Gemini features and was expanding its AI agent capabilities. The rapid deployment of models like Gemini Decision 1, tailored for specific probabilistic reasoning, points to a strategy of breaking down AI problems into narrower, commercially viable applications.

Hardware Investments Support AI Ambitions

Google’s commitment extends beyond software. The company invests heavily in the underlying infrastructure that powers these models. On November 10, Google broadened its AI hardware portfolio with two new specialized Tensor Processing Units (TPUs). These are designed for confidential model training and increased memory capacity, as reported by TechCrunch. These chips build on earlier infrastructure announcements, including the open-sourcing of the LLM-D 1.0 distributed inference stack and TorchTPU for native PyTorch integration on October 21.

The most ambitious infrastructure play came on October 1. Google launched its first Suncatcher orbital data center test, an initiative pushing computing into space. These hardware advancements provide the critical processing power and specialized capabilities for large-scale AI training and inference. They make the rollout of increasingly complex Gemini models possible.

Expanding AI to New Products and Services

Beyond foundational models and core infrastructure, Google embeds AI into a wider array of products. The company launched the Universal Cart shopping tool in the US on October 16. This is a consumer-facing application of its AI capabilities. More recently, Google revealed Tabletop, an experimental AI tool. Tabletop generates customized collaborative hubs to help users coordinate. Users can “vibecode personalized interactive tools” by describing their needs, according to The Verge. This move into collaboration apps shows Google’s efforts to integrate AI into daily user workflows. Our September 8 coverage noted this strategy when detailing Gemini’s deepening integration across Pixel devices and services.

Competition for AI tools continues to heat up. OpenAI, for example, recently rolled out new business profiles. It aims to make ChatGPT a stronger consumer discovery engine. This draws comparisons to Google’s traditional strengths in information access. Both major players seek to expand their reach across enterprise and consumer applications.

Data Ethics Remain a Concern

Real-world data handling continues to present challenges amidst rapid technological advancements. In a separate development reported by Wired, the union representing flight attendants for the defunct Spirit Airlines struck a deal with Google. The deal scrubbed sensitive employee records from a $10 million data sale. This incident, while distinct from AI model development, shows the ongoing challenge of data privacy and ethical considerations as companies like Google acquire and process vast amounts of information. The initial restrictions on Gemini 4 Argon for safety concerns also show Google’s awareness of managing powerful AI systems carefully.

What is Google’s Gemini Decision 1 API?

The Gemini Decision 1 API is a new Google model designed to return calibrated probabilities for up to 1,000 yes/no, choice, and rubric questions across various input types like text, images, audio, video, and PDFs.

How much does the Gemini Decision 1 API cost?

The Gemini Decision 1 API is priced at $0.10 per 1 million input tokens.

What other Gemini 4 models has Google released recently?

Google launched Gemini 4 Argon on October 1, followed by Gemini 4 Carbon on October 29, which focuses on cybersecurity and coding tasks and has since opened to developers and enterprises.

Compiled by Launch91 Desk from the sources linked above. More about Launch91.