DeepRAP: EIC Pathfinder 2026 grant for cognitive AI

The European Innovation Council has opened applications for DeepRAP: Deep Reasoning, Abstraction & Planning towards trustworthy Cognitive AI Systems. The topic code is HORIZON-EIC-2026-PATHFINDERCHALLENGES-01-03, and it sits inside EIC Pathfinder Challenges 2026 under Horizon Europe. The submission session opened on 22 July 2026, so teams now have until 28 October 2026, 17:00 Brussels time.

DeepRAP funds research that gives AI systems deeper reasoning, abstraction and planning. Because the EIC runs three Pathfinder Challenges this year on a shared budget of EUR 96 million, each challenge holds roughly EUR 32 million. Under DeepRAP the EIC expects to select 8 projects, and a single project can request up to EUR 4 million at a 100% funding rate.

What DeepRAP funds

Generative models read patterns well and answer accurately across a wide range of tasks. Still, they fail often on logic problems and long-horizon planning, even when a provably correct solution exists. That gap limits their use wherever an error costs a lot: medicine, cybersecurity, critical infrastructure control.

So the EIC asks applicants to move past both symbolic methods (rules, decision trees, symbolic regression) and neural ones (deep learning, large language models, reinforcement learning). Teams may combine existing techniques, for example neuro-symbolic AI, or build entirely new frameworks. Ideas from neuroscience, biology, physics or philosophy also fit.

Each proposal must cover at least one of the three cognitive capabilities below. Projects that cover several capabilities and explain how they join them carry more weight during portfolio building.

Deep reasoning

This strand covers causal inference, logical and probabilistic reasoning, context-aware and commonsense decisions, end-to-end differentiable reasoning, and multimodal reasoning across vision, space and time. A system should show why a pattern appears, name the causes behind it, and give reasons a person can check. Since the EIC wants accountable behaviour in demanding settings, it specifically encourages neuro-symbolic designs that join neural learning with structured symbolic inference.

Deep abstraction

Here the EIC looks for generalisation from small data: concept formation, analogical reasoning, internal world models, semantic understanding of context, and transfer learning. Because real tasks rarely come with large clean datasets, moving a learned concept into a new domain matters more than the size of the training set.

Deep planning

This strand covers planning algorithms for open worlds, agentic systems and real-time uncertainty. Specifically, the call names hierarchical planning across several time horizons, contingency planning with fallback strategies, and continual re-planning as conditions change. Formal guarantees and explainable decisions give a proposal an edge, so teams that add verification methods start ahead.

What results the EIC expects

The work programme names the expected outcomes in detail. Therefore a proposal has to show how the project reaches each of them.

• Models or architectures that handle multimodal data and knowledge, account for uncertainty, and train under limited computing resources.
• Provable trustworthiness mechanisms: explainability, openness, fairness, risk evaluation, security, and compliance with fundamental rights and the EU AI Act.
• A demonstration of the capabilities inside a cognitive AI system at TRL 4 on real tasks such as scientific discovery, decision support and problem solving, plus simulations at scale.
• New methods and metrics for evaluating and certifying reasoning, trustworthiness and the use of computing resources.
• FAIR principles, so all data, models and results stay findable, accessible, interoperable and reusable.
• Shared work with EU initiatives: TEFs, EBRAINS, RAISE, the AI-on-Demand Platform and the Quantum Flagship.

TRL 4 means a technology validated in the lab. Because that is the ceiling for Pathfinder, promises of a market-ready product weaken a proposal.

Who can apply, and where Ukraine stands

Pathfinder Challenges accept proposals from consortia and from single legal entities established in an EU Member State or an Associated Country. That rule separates Challenges from Pathfinder Open, where a consortium is mandatory.

• Two-partner consortium: two independent legal entities from two different Member States or Associated Countries.
• Consortium of three or more: at least three independent legal entities, each in a different country, and at least one of them in a Member State.
• Single applicant: universities, research organisations, SMEs, start-ups and natural persons all qualify. Mid-caps and larger companies cannot apply on their own.

Ukraine takes part in Horizon Europe as an Associated Country. As a result, Ukrainian universities, research institutes and deep-tech companies apply on the same terms as EU teams, and they can coordinate a project. First check your PIC and legal entity validation, because a missing LEAR appointment slows everything down at the worst moment.

One restriction applies to proposals that touch the evolution of European communication networks, including 5G and later technologies. Annex II, Section B1 of the EIC Work Programme 2026 sets out the details.

Sections 1 to 3 of Part B run to a maximum of 30 A4 pages. The submission system hides anything beyond that limit, so evaluators never read the extra pages.

How portfolio building shapes your proposal

Evaluation runs in two steps. First, expert evaluators score each proposal separately against Excellence, Impact, and Quality and Efficiency of the Implementation. After that, the committee maps every above-threshold proposal into categories and assembles a portfolio from them.

There are four categories: cognitive capability (reasoning, abstraction, planning or a combination), technological approach (neuro-symbolic AI, deep learning, reinforcement learning, cognitive architectures, interdisciplinary frameworks, formal methods, multimodality, trustworthiness mechanisms), application domain (scientific discovery, decision support, autonomous systems, human-AI collaboration, cybersecurity, industry, societal applications) and shared portfolio work (benchmarks, interoperability, joint pilots).

One detail changes how you prepare. A highly ranked proposal can still miss the portfolio if the committee sees no common ground with other projects. Equally, a proposal too close to one already selected may drop out. So the project has to stand apart and stay usable for joint work at the same time.

The Challenge Guide

The Challenge Guide asks for two things inside the proposal text. First, a self-assessment table where the team maps its own project to the categories and justifies each choice. Second, a dedicated work package for portfolio activities with at least 10 person-months, using the template in Annex 1 of the guide. If the package is missing, evaluators will not cut the score, though during grant preparation the EIC will ask you to add it, and the maximum grant stays the same.

Once projects start, they work through four groups: technology integration and demonstration, transition of technology to innovation, regulation and ethics, and communication. Online meetings happen roughly every 3 months, plus one annual meeting in person. Programme Manager Hedi (Mohamed) Karray coordinates this work and builds the portfolio strategic plan together with the consortia.

Funding terms and what follows the grant

The EIC pays the grant as a lump sum under the Lump Sum MGA at a 100% rate on eligible costs. The indicative size reaches EUR 4 million, and you can request more if the detailed budget table justifies it. Applicants hear the outcome roughly 5 months after the deadline, while grant agreements follow within about 8 months.

Winners also reach a second layer of support: booster grants up to EUR 50,000, Business Acceleration Services, a fast track to the EIC Accelerator after a project review, and the option to apply for EIC Transition once the project reaches TRL 3 or 4.

Competition runs hard. In the 2025 Challenges round the committee received 647 eligible proposals and funded 4.6% of them. Because of that, fit with the portfolio categories weighs as much as the science itself: a strong idea with an unclear place in the portfolio loses at step two.

You submit through the Funding & Tenders Portal on the DeepRAP topic page. The same page holds the application template, the detailed budget table and the Challenge Guide with the portfolio categories. The EIC scheduled an info session on this challenge for 30 March 2026 and publishes the recording in the events section of its website.