Planet AI’s winning formula – European AI expertise on the global stage.
At the end of June, Planet AI, the Bechtle Group’s specialist for artificial intelligence, was officially confirmed as the winner of the world’s most demanding competition for document-based AI systems. The challenge was every bit as complex as its title suggests: the ICDAR 2026 Competition on Multimodal Reasoning over Documents in Multiple Domains, or DocVQA 2026 for short, required AI systems to demonstrate a deep understanding of a wide range of document types. Against a strong international field, Planet AI secured overall victory.
What was the challenge?
At DocVQA 2026, international teams competed with AI systems built on the latest language models from Google, Anthropic, OpenAI and Alibaba. The task was to understand documents and answer questions that required complex reasoning across text, tables, charts and document layouts to arrive at the correct answer. With eight document types ranging from financial reports to comic books, DocVQA 2026 raised the bar higher than ever before.
How did Planet AI stand out?
Rather than relying on a single model, Planet AI took an agentic approach. Its Distributed Cognitive Architecture (DCA) brings together multiple specialised models that analyse documents in parallel and combine their findings into a consensus view. The system is further strengthened by IDA, Planet AI’s proprietary document analysis technology, which provides the structured foundation for understanding and processing content. The outcome speaks for itself: Planet AI took first place overall and won seven of the competition’s eight document categories.
Was this Planet AI’s first international success?
DocVQA 2026 marked Planet AI’s first participation in the competition. The underlying technology, however, has a long track record. Between 2014 and 2019, Planet AI won seven international competitions in the field of document analysis. The latest success brings that expertise to a new level of global visibility.
What’s next?
The official award ceremony will take place at this year’s ICDAR conference in Vienna from 30 August to 4 September 2026. For Planet AI, the result highlights the strength of artificial intelligence made in Europe. The focus now shifts to applying these capabilities in real-world projects, particularly where requirements such as data sovereignty and regulatory compliance play a critical role.
Is the approach ready for real-world use?
The technology is clearly designed for production use. Systems of this kind can analyse a wide range of documents and identify relationships across them, from invoices and contracts to complex technical documentation. For businesses and public sector organisations alike, this can mean more automation, fewer errors and faster decision-making.
DocVQA 2026 at a glance.
- Organiser: Computer Vision Center (CVC), Universitat Autònoma de Barcelona
- Participants: International teams from academia and industry.
- Challenge: AI systems answer questions about real-world documents across eight document categories
- 2026: The most demanding edition of the competition to date.
- Winner (35+ billion parameter category): Planet AI GmbH, Rostock (competing in DocVQA for the first time)
- Winning margin: First place in seven of eight categories; overall winner by around 20 percentage points ahead of the standard mixture-of-experts baseline and nearly 14 percentage points ahead of the runner-up team