Gemini API adds Webhooks: Google solves the pain point of long-task polling, and Batch/Veo can be pushed instantly

Google launched the Webhooks feature on the Gemini API on May 4, addressing developers’ pain points with long-running jobs. In a post on the official Google blog, Google explained that Webhooks are an event-driven push notification mechanism, so developers no longer need to continuously poll the Gemini API for task status. After the API completes, it will proactively push the results to the endpoint specified by the developer. Logan Kilpatrick (@OfficialLoganK), Head of Developer Relations for Google AI, said on X that this is an “important step for long-running job DevX.”

What problem it solves: the cost of polling for batch, video generation, and long reasoning

In the past, Gemini API developers handling batch (batch processing), video generation (Veo 2), and long reasoning tasks had to call the status endpoint every few seconds to check task progress. This approach is not ideal in terms of resource consumption, API quota, and latency:

Resource waste—many pointless status check calls, consuming API quota

Latency not controllable—if the polling interval is too short, it burns quota; if too long, the result is known later

Code complexity—clients must implement a state machine to manage polling for multiple concurrent tasks

Webhooks flip this pattern: developers register a callback URL, and when Gemini completes the task it proactively POSTs the result to that URL, so the client only needs to handle the push.

Applicable scenarios: Batch API, Veo 2 videos, and long-context reasoning

This Webhooks release mainly applies to three types of asynchronous tasks:

Batch API—Gemini’s batch processing endpoint for large-scale text, embedding vectors, and classification tasks. The official provides a 50% discount, targets a 24-hour response time, and in practice it usually completes within a few hours

Video generation (Veo 2)—generating a single video takes minutes, and previously developers had to keep polling

Long context reasoning—analysis of long documents with 1M tokens or more, where Gemini’s internal processing may take tens of seconds to several minutes

From a developer implementation standpoint, after registering a webhook, the client can “submit the job and forget it,” and when the result is ready Gemini naturally notifies the client. This model is especially suitable for serverless architectures: the backend only needs to be awakened when events arrive, without maintaining a polling process.

Side-by-side with OpenAI and Anthropic: who did what first, who did what next

Long-running job webhooks progress across three major AI platforms:

Google Gemini: Webhooks released on May 4 (this case), covering batch, videos, and long reasoning

OpenAI: Long-running jobs such as Codex and Sora 2 currently rely mainly on SSE (Server-Sent Events) streaming. Batch tasks have a dedicated endpoint, but there is no native webhook

Anthropic: Claude API has no native webhooks; Claude Code uses a polling mechanism internally to handle long-running tasks

On the DevX (developer experience) axis, Google has clearly increased investment over the past 12 months—from the 1M context of Gemini 2.5 Pro, the visualized development experience of AI Studio, the Agent Designer and Memory Bank introduced at Cloud Next 2026, to this Webhooks release. Compared with OpenAI’s priority on “direct consumer-facing products” (ChatGPT, Operator), Google is taking a “enterprise/developer infrastructure” route. Webhooks are a concrete piece of that strategy.

Next to watch: webhook security mechanisms, and the scope of supported models

Key focus areas for the next stage:

Webhook security mechanisms—whether Gemini provides HMAC signature verification to prevent forged requests from overwhelming the callback URL

Model scope expansion—currently covers batch, Veo 2, and long reasoning. Will it also support future capabilities such as Imagen image generation, Speech-to-Speech, and whether Gemini Live is included

OpenAI and Anthropic responses—once Google brings DevX to this level, whether competitors follow suit

For practical use by Taiwan developers: if you’re using the Gemini API for batch jobs (for example, batch processing customer data classification and document summarization), Webhooks are a feature worth integrating immediately, as they can significantly reduce API quota usage and system complexity.

This article, Gemini API Pushes Webhooks: Google Solves Long-Running Job Polling Pain Points, with Batch/Veo Supporting Real-Time Push, first appeared on Lianxin News ABMedia.