Google Forces Vertex AI Migration as Sora 2 Launches: Week of 29 September 2025

Google is forcing a major migration for Vertex AI users whilst OpenAI steals headlines with Sora 2's launch. With 42 signals this week including three critical updates, the AI landscape is shifting rapidly beneath developers' feet.

The Big Moves

Google's Vertex AI Overhaul Demands Immediate Attention

Google has announced sweeping changes to Vertex AI that will impact thousands of developers by June 2026. The company is deprecating multiple image and video generation endpoints whilst simultaneously rolling out new capabilities including RAG Cross Corpus Retrieval, the Lyria 3 model, and Vector Search 2.0.

The deprecation timeline is unforgiving. Users relying on the current image and video generation endpoints have until 30 June 2026 to migrate to recommended replacements, or face complete service disruption. This isn't a gentle nudge towards newer APIs; it's a forced march that will require substantial development effort for anyone building on these endpoints.

Meanwhile, Google is sweetening the deal with genuinely useful additions. The new C# Generative AI SDK brings GenerateContentAsync, GenerateContentStreamAsync, and GenerateImagesAsync capabilities to .NET developers. Vertex AI Workbench v2 is getting a complete refresh with Debian 12, Python 3.12, and Workforce Identity Federation support, though it's dropping JupyterLab 3, TensorFlow, and PyTorch support in the process.

The strategic message is clear: Google wants to consolidate its AI offerings around newer, more capable endpoints whilst forcing users off legacy systems. For developers, this means immediate planning for migration paths and potential application rewrites.

OpenAI's Sora 2 Redefines Content Generation

OpenAI has launched Sora 2 with advanced video and audio generation capabilities that represent a genuine leap forward in AI-generated content. The new model delivers realistic physics simulation and multilingual support, backed by a dedicated iOS app for immediate experimentation and social sharing.

The initial rollout targets the US and Canada with free usage limits, positioning Sora 2 as both a consumer product and a developer platform. This dual approach suggests OpenAI is serious about mainstream adoption whilst building the foundation for enterprise integration.

What makes Sora 2 particularly significant is its positioning as general-purpose simulation technology rather than just another content generation tool. The realistic physics and multilingual capabilities suggest applications far beyond social media content, potentially disrupting industries from education to entertainment.

For developers, Sora 2 represents both opportunity and competition. The iOS app lowers the barrier to experimentation, but the underlying technology could reshape how we think about automated content creation across multiple industries.

Qdrant Addresses Critical Data Corruption Issues

Qdrant has released emergency fixes for critical data corruption and deadlock issues that could severely impact vector search reliability. Version 1.15.5 addresses data corruption during snapshots, segment corruption, and deadlocks in REST operations.

These aren't minor bug fixes; they're critical stability improvements that could prevent data loss and service outages. For organisations running large vector databases, immediate updates are essential to maintain data integrity and avoid potential disasters.

The timing is particularly concerning given the increasing reliance on vector databases for RAG applications and AI-powered search. Any data corruption in these systems could cascade through dependent applications, making this update non-negotiable for production environments.

Worth Watching

Azure OpenAI Expands Audio Capabilities

Microsoft has released the GPT-4o Audio Transcription and Diarization model, offering real-time transcription across 100+ languages with speaker identification. The model's low latency makes it ideal for customer support and virtual meetings, transforming voice data into structured insights. The addition of SIP support for the Realtime API enables direct telephony integration, opening new use cases for voice-first applications.

AWS Bedrock Grows Model Selection

AWS continues expanding Bedrock's model ecosystem with batch inference support for five new DeepSeek and Qwen models. Cohere Embed v4 is now available, providing improved embedding quality for vector search applications. Amazon Bedrock Guardrails has expanded to Asia Pacific (Melbourne) with cross-region inference capabilities, offering better geographic distribution and resilience.

Elastic Gains Industry Recognition

Elastic has been named a Leader in The Forrester Wave for Cognitive Search Platforms, validating its continued innovation in AI-powered search. The recognition highlights Elasticsearch's strengths in scalability, customisation, and modern AI search capabilities including ES|QL and serverless deployments.

Perplexity Launches Developer Tools

Perplexity has released official Python and TypeScript SDKs alongside enhanced Search API features including language preferences and domain filtering. The interactive playground removes API key requirements for initial testing, accelerating developer onboarding and experimentation.

Amazon OpenSearch Adds ML Batch Processing

Amazon OpenSearch Ingestion now supports ML offline batch inference, enabling efficient asynchronous data enrichment using Amazon Bedrock and SageMaker models. This capability provides a cost-effective solution for processing large datasets at scale.

Quick Hits

• Google deprecates MedLM model: Access ends 29 September 2025, requiring migration to Gemini 2.5 Pro or alternatives
• Azure OpenAI adds PII detection: Built-in content filter automatically identifies and blocks sensitive information
• OpenAI releases GPT-image-1-mini: Smaller, more cost-effective image generation model for global deployment
• Google adds DeepSeek-V3.2-Exp: New model available in Vertex AI Model Garden
• OpenAI system uptime: APIs (99.06%), ChatGPT (98.65%), Sora (99.84%) for October 2025-January 2026 period

The Week Ahead

The immediate priority is assessing Google's Vertex AI changes and their impact on your applications. The June 2026 deadline might seem distant, but migration planning should begin immediately given the scope of affected endpoints.

Watch for OpenAI's expansion of Sora 2 beyond the initial US/Canada rollout. The iOS app's performance and user adoption will signal whether OpenAI can successfully bridge consumer and enterprise markets.

Qdrant users should prioritise the v1.15.5 update, particularly those running large clusters or high-volume operations where data corruption could have cascading effects.

Expect continued model launches as providers race to expand their offerings before year-end. The pattern of simultaneous capability expansion and legacy deprecation is becoming the industry standard, requiring constant vigilance from development teams.