Google Adds Batch Mode to Gemini API for Cheaper, Scalable AI Jobs

Google has introduced a Batch Mode for the Gemini API, aimed at developers running large-scale, non-latency-sensitive workloads. The new endpoint supports asynchronous jobs with up to 24-hour turnaround, bundling multiple prompts into a single API call. Users can upload JSONL files up to 2 GB in size, with a cap of 2,000 total requests per job. Batch Mode also supports tool integration like Google Search and context caching to improve efficiency across large datasets.

Pricing for Batch Mode is set at 50 % less than the standard synchronous API, making it a cost-efficient option for teams focused on content generation, model evaluation, or data labeling. Google emphasizes the scalability of this approach, offering higher rate limits and reduced overhead for managing large volumes of calls. This positions Gemini more competitively against alternatives like OpenAI’s batch processing and aligns with broader enterprise use cases.

The update also signals Google’s intent to make its GenAI infrastructure more developer-friendly and operationally practical. By targeting workflows where real-time speed is unnecessary, Batch Mode unlocks more budget-conscious applications of Gemini models. It is currently available in public preview for Gemini 1.0 Pro and 1.5 Pro across Vertex AI and Google AI Studio.