Microsoft Germany leaks GPT-4 launch next week with multimodal support

Just as the world was getting used to text-only prompts, Microsoft Germany practically spilled the beans on OpenAI's next move.

Speaking at an "AI in Focus - Digital Kickoff" event on March 9, Microsoft Germany CTO Andreas Braun casually dropped that GPT-4 is launching next week. And the headline feature is multimodality, which will unlock entirely new possibilities, including video.

GPT-4 goes multimodal

The event was held in German for pre-registered corporate attendees, mostly focusing on enterprise use cases for Azure-OpenAI. But Braun's offhand remark immediately stole the show.

Braun called large language models a genuine game changer because they teach machines to statistically understand natural language that was previously only readable by humans. He noted the technology now fundamentally works across all languages, allowing someone to ask a question in German and receive a reply in Italian. With multimodality, Microsoft and OpenAI aim to make the models truly comprehensive.

Holger Kenn, Chief Technologist for Business Development AI at Microsoft Germany, added some technical color, explaining that multimodal AI can translate text into images, music, and video. Given Microsoft's recent release of its Kosmos-1 model in early March, the leap into multimodal territory was already brewing, but Braun's confirmation puts a very tight timeline on it.

Real enterprise hours on the line

Microsoft Germany CEO Marianne Janik compared the current AI wave to an "iPhone moment," arguing that the goal is not eliminating jobs, but changing how repetitive work gets done. She also emphasized that Microsoft does not use enterprise customer data to train its models.

To show what is already happening, Senior AI Specialist Clemens Siebler shared a Dutch enterprise case study. The client receives 30,000 calls a day; by using speech-to-text and AI summarization instead of manual call logging, the company saves 500 working hours daily. Siebler noted that a single developer built the core implementation in just two weeks after a two-hour prototype.

Addressing hallucination risks, Siebler acknowledged that models do not always answer correctly. Microsoft is building confidence metrics and relying on iterative thumbs-up and thumbs-down feedback loops, especially when customers query internal knowledge bases.

Why it matters

Bringing multimodality into GPT-4 changes the entire AI playground overnight. For developers and creators, handling text, images, and video inside a single unified framework will collapse the friction of building multimodal media tools. That said, businesses diving in should stay realistic: as Siebler pointed out, verifying generative outputs remains messy, so pairing models with internal validation workflows is still essential before putting them in front of end users.

For the broader industry, OpenAI and Microsoft are essentially shifting the goalposts before rivals have even matched GPT-3.5. Competing only on text benchmarks looks increasingly dated when the frontier is expanding into video and cross-modal reasoning.

Strategically, floating the announcement at a regional German enterprise event is a clever nudge to conservative European companies that are historically slow to replace legacy systems. However, with the EU AI Act looming and threatening massive compliance penalties, pushing ambitious multimodal tech into regulated enterprise workflows may encounter plenty of friction down the road.