Google Forces Vertex AI Migration as Critical Endpoints Face Sunset

Google dropped a migration bombshell this week: multiple Imagen and Veo generation endpoints are being deprecated across Vertex AI, with a hard sunset date of 30 June 2026. This isn't a gentle nudge towards newer APIs—it's a forced march that will break existing integrations if teams don't act. Meanwhile, OpenAI launched GPT-5.2 Pro with enhanced professional capabilities, and Together AI released a breaking SDK update that demands immediate attention.

What's changing with Vertex AI endpoints?

Google's Vertex AI deprecation affects multiple critical services across image and video generation. The deprecated endpoints include several Imagen variants and Veo generation APIs that many teams have built into production workflows. What makes this particularly challenging is the 18-month migration window—generous in theory, but potentially problematic given the scope of changes required.

The timing suggests Google is consolidating its generative AI offerings around newer, more efficient architectures. The company is simultaneously releasing Gemini 2.5 Flash with native audio capabilities and updating Vertex AI Workbench to version 4+, indicating a broader platform modernisation. Teams using the deprecated endpoints need to start migration planning now, as the June 2026 deadline will arrive faster than expected once you factor in testing, validation, and deployment cycles.

The migration path isn't straightforward either. Google hasn't provided direct one-to-one replacements for all deprecated functionality, meaning some teams may need to redesign their image and video generation workflows entirely. This could be an opportunity to modernise, but it's also a forced investment that many hadn't budgeted for.

Together AI's SDK overhaul demands immediate attention

Together AI's Python SDK v2.0 represents the kind of breaking change that keeps DevOps teams awake at night. The release candidate introduces significant architectural modifications that will require code updates across existing integrations. With a one-month RC period before the final release, teams have a narrow window to test and adapt.

The SDK overhaul reflects Together AI's maturation as a platform, but the breaking changes are substantial enough to warrant dedicated sprint time. Any production systems using the current SDK will need updates to maintain functionality after the v2.0 release. The timing is particularly challenging for teams already managing year-end releases and planning for 2026.

What's encouraging is that Together AI is providing a proper release candidate period rather than dropping breaking changes without warning. Teams should use this window to identify affected code paths and plan migration work. The architectural improvements in v2.0 suggest better performance and reliability, but only if you complete the migration successfully.

OpenAI's GPT-5.2 Pro targets professional workflows

OpenAI's GPT-5.2 Pro launch focuses squarely on professional knowledge work, with improvements in spreadsheet analysis, code generation, and long-context processing. The model represents a clear evolution towards enterprise use cases, offering better performance for the complex, multi-step workflows that define modern knowledge work.

The emphasis on professional capabilities suggests OpenAI is responding to enterprise feedback about real-world AI deployment challenges. Improvements in cost-effectiveness and speed address two of the biggest barriers to production AI adoption. The enhanced code generation capabilities could be particularly significant for development teams looking to integrate AI into their workflows without compromising quality or security.

Worth watching this week

Anthropic's enterprise push gains momentum with the expanded Accenture partnership aimed at moving companies from AI pilots to production deployments. The focus on training 30,000 Accenture professionals on Claude models creates a substantial ecosystem of practitioners, particularly in regulated industries like financial services and healthcare.

Mistral AI's model expansion continues with the release of Devstral 2, Devstral Small 2, and Mistral Vibe. These releases expand Mistral's offerings across code generation and general use cases, providing more options for teams evaluating alternatives to OpenAI and Anthropic models.

Hugging Face's infrastructure improvements include the launch of HuggingChat for Docs and enhanced Team & Enterprise content organisation. These changes strengthen Hugging Face's position as a comprehensive platform for AI development, particularly for teams building on open-source models.

Weaviate's stability focus shows in multiple releases (v1.32.22, v1.32.23, v1.33.10) emphasising authentication improvements, OIDC enhancements, and backup reliability. The consistent focus on infrastructure stability suggests Weaviate is preparing for increased enterprise adoption.

Cohere's Rerank 4.0 introduces a new foundational model for ranking tasks, representing an upgrade to their existing capabilities. Teams heavily reliant on ranking should evaluate whether the improved performance justifies migration from existing solutions.

Quick hits

• AWS Bedrock adds API keys for GovCloud development, streamlining access for government and defence applications • Disney expands Sora integration with Disney IP, unlocking new creative possibilities for Disney+ users • Anthropic donates Model Context Protocol to the Agentic AI Foundation, establishing it as a vendor-neutral open standard • Pinecone adds GPT-5 support and dedicated read nodes alongside new annual pricing options • llama.cpp introduces dynamic model management for efficient switching between multiple models without server restarts

The week ahead

The Vertex AI deprecation timeline means teams should begin migration planning immediately, even with the June 2026 deadline. Together AI's SDK v2.0 final release is imminent, so complete your testing and migration work during the RC period. OpenAI's GPT-5.2 Pro warrants evaluation for teams focused on professional knowledge work applications.

Watch for additional details on Google's migration paths for deprecated endpoints—the current guidance lacks specificity that production teams need. The broader trend towards platform consolidation and breaking changes suggests 2026 will be a year of significant infrastructure updates across multiple AI providers.