Frontier AI Benchmarking Datasets 2026: Historical Innovate UK Competition for UK AI Benchmark and Dataset Projects
Innovate UK ran a closed 2026 competition for collaborations creating benchmark datasets, curated annotated datasets, and evaluation harnesses for priority AI missions in health, life sciences, and advanced materials.
Frontier AI Benchmarking Datasets 2026: Historical Innovate UK Competition for UK AI Benchmark and Dataset Projects
This is a historical reference for a closed Innovate UK competition. The official UKRI opportunity page marks Frontier AI Benchmarking Datasets as closed, and the Innovation Funding Service says that the competition is now closed. The official pages checked for this update show only this 2026 round; no later cycle is announced on those pages. Do not treat this page as an open application route or as evidence that a new round has been approved.
At a glance
| Detail | Information |
|---|---|
| Opportunity | Frontier AI Benchmarking Datasets |
| Funder | Innovate UK, part of UK Research and Innovation |
| Funding type | Grant |
| Competition allocation | Share of up to £4.5 million; Innovate UK described a minimum £4.5 million allocation |
| Competition status | Closed; historical reference |
| Opening date | 2026-04-21 |
| Closing date used for this archive | 2026-05-29 at 11:00am UK time |
| Project size | £500,000 to £750,000 total eligible costs |
| Project length | 6 to 12 months |
| Start date | By 2026-09-01 |
| End date | By 2027-08-31 |
| Lead organisation | UK registered business of any size or RTO |
| Consortium rule | At least one UK-registered SME claiming grant funding |
| Official source | UKRI opportunity page |
| Application portal | Innovation Funding Service competition brief |
What this opportunity is trying to buy
This competition was not just about generic AI research. The official brief was specific: Innovate UK wanted projects that created high-quality benchmark datasets, representative dataset slices, and evaluation harnesses that could help the UK evaluate new AI models while developing larger AI-ready annotated and curated datasets.
The first mission was AI-enabled health and life sciences, including medicines discovery, medicines development and manufacturing process optimisation, predictive healthcare applications, and clinical trials. Genomics and multi-omics could be enabling technologies across those priorities. The second mission was advanced materials with AI, covering material prediction, physics machine-learning models for discovery and simulation acceleration, and multimodal knowledge discovery platforms in areas such as aerospace, net zero technologies, defence materials, and semiconductors. The common thread was a benchmark or dataset that could support a practical AI use case and make model performance comparable against a documented reference set.
That focus mattered because a broad AI data project would not meet the brief. The required package was an open benchmark package with a task definition and evaluation protocol, an openly accessible benchmark dataset, a complete evaluation harness usable by third parties, documentation and metadata, a fully curated and annotated full dataset, and clear IP licensing and access terms for the full dataset slice. A slide deck, concept note, or vague data platform would not have met those deliverables.
The Innovation Funding Service says Innovate UK would invest a minimum of £4.5 million, subject to receiving enough high-quality applications, while describing the opportunity as a share of up to £4.5 million. It also warns that the competition is selective and that similar competitions had around a 10% chance of success. Scope therefore mattered alongside evidence that the consortium could deliver a credible benchmark product within 6 to 12 months.
Who should pay attention
The best fit was a consortium that already had access to valuable data and could combine that access with data engineering, annotation, governance, and benchmark design. The official competition text encouraged consortia that brought together data-owning organisations and partners with expertise in data engineering, annotation, and benchmarking.
That usually meant a mix like this:
- a UK business or RTO that could lead the grant,
- one or more partners that own or control the underlying data,
- a technical team that can clean, annotate, and structure the data,
- and an organisation that understood how the resulting benchmark could be used commercially or by third parties.
The call was a strong fit for teams that could answer a simple question: what dataset or benchmark would make AI evaluation materially better in a priority sector, and why could the market not already do that well enough?
It was less suitable for:
- solo applicants,
- pure academic-led proposals where a university wants to lead,
- projects that needed more than 12 months to become useful,
- or projects that did not have a clear path to a benchmark package and usable dataset release.
The lead requirement was especially important. The official brief said the lead had to be a UK registered business of any size or an RTO. If an RTO led, it had to collaborate with two businesses: one SME and one business of any size. Academic institutions could collaborate, but could not lead. A consortium also had to contain at least one UK registered SME claiming grant funding.
Eligibility and project rules
The published rules were concise but strict. A project had to:
- be a collaboration only,
- have total eligible costs between £500,000 and £750,000,
- last 6 to 12 months,
- start by 1 September 2026,
- end by 31 August 2027,
- and be carried out in the UK with exploitation from or within the UK.
The consortium also had to include at least one UK-registered SME claiming grant funding on the application. That was a structural requirement, not an optional adviser role. To count as an eligible collaboration, the lead and at least one other organisation had to apply for funding, explain the collaboration structure, and ensure that no one partner accounted for more than 70% of total eligible costs.
Another practical rule concerned cost concentration. No one partner could account for more than 70% of total eligible costs. The project also had to include only eligible costs, and subcontractor costs were capped at 20% of total project costs. Those rules pushed the proposal toward a real partnership rather than a single dominant contractor with minor advisers attached.
If the project used health and life sciences data, the brief required appropriate data governance and privacy protections, and any released data had to be anonymised or de-identified. That required more than a generic ethics statement: the consortium needed a defensible plan for rights, consent where relevant, access control, and release. Non-funded partners were permitted, including non-UK partners, but the official rules distinguished their status from funded organisations and their UK work and exploitation obligations.
The opportunity also had an accessibility note: applicants could request reasonable adjustments, and Innovate UK recommended contacting it at least 15 working days before the close date if support was needed. That deadline has passed with the competition, but the contact route remains useful if Innovate UK announces a replacement competition.
What reviewers are likely to expect
The competition brief spelled out the deliverables. A strong application needed to be built around all of them, not just one.
Applicants needed to show that the project would deliver:
- an open benchmark package, including the task definition and evaluation protocol,
- an openly accessible benchmark dataset,
- a complete evaluation harness usable by third parties,
- documentation and metadata,
- a fully curated and annotated full dataset,
- and clear details on the intellectual property licence and access route for the full dataset slice.
That list showed what the funder meant by a complete output. A proposal that only promised a data release without evaluation tooling was incomplete. A proposal that only promised an internal benchmark without access terms was also incomplete. Innovate UK wanted a usable benchmark system, not just a data dump.
The brief also said strong applications would clearly demonstrate:
- the value added by the benchmark or dataset over existing resources,
- the industry opportunity the work enables,
- and how the project improves evaluation or training for AI and machine learning models.
That meant the written case needed to describe more than the data. It needed to describe the performance gap: what the field lacked, why that gap mattered, and what new measurement or training capability would become possible if the benchmark existed.
If you are proposing a health-related benchmark, the reviewer will likely want to see stronger data governance detail and a sharper explanation of why the release is safe and useful. If you are proposing an advanced materials benchmark, the emphasis may shift more toward industrial relevance, model validation, and commercial access.
Timeline and deadline details
The published pages record a very tight window for the closed round:
- the UKRI opportunity page lists the opening date as 21 April 2026,
- the Innovation Funding Service competition brief also lists 21 April 2026,
- the Innovation Funding Service page lists the operational closing date as 29 May 2026 at 11:00am UK time,
- the UKRI landing page currently displays 27 May 2026 at 11:00am UK time, creating a date discrepancy that should be retained in an archive rather than silently rewritten,
- and the project had to start by 1 September 2026 and end by 31 August 2027.
Because the application was submitted through the Innovation Funding Service, this page uses 29 May 2026 as the archived deadline in front matter and in the summary table. The UKRI page is still useful as the funder landing page, but its displayed 27 May date should be treated as a conflicting official display rather than evidence of a new round. Both official pages show the competition as closed.
Before the round closed, Innovate UK recommended that applicants needing accessibility support or reasonable adjustments contact it at least 15 working days before the closing date. That advice belonged to the 2026 application window and cannot reopen this competition.
How to frame the application
Because the project size was only 6 to 12 months, the application needed to read like a short, executable product plan. The strongest structure for that round was:
1. Define the benchmark problem precisely
Applicants were expected to start with a concrete use case, not a domain slogan. The proposal needed to explain whether it benchmarked model performance on a specific drug-discovery task, clinical workflow, materials-characterisation task, or manufacturing problem. Assessors needed to understand what the model would be judged on.
2. Show why your data is representative
The brief refers to representative dataset slices. That phrase matters. It implies that the benchmark should reflect a real subset of a bigger data world, not a toy sample chosen only because it was easy to curate.
The application needed to describe why the slice was representative, what it excluded, and what it preserved.
3. Explain your annotation and curation plan
The call explicitly asked for curated and annotated datasets. Annotation quality was therefore not an afterthought. An application needed to spell out:
- who will annotate,
- what the label taxonomy is,
- how disagreements will be resolved,
- how quality assurance will work,
- and how you will document the dataset for external users.
4. Build the evaluation harness as a product
The competition wanted a harness usable by third parties. The benchmark therefore needed to be reproducible and clear enough that another team could run it and compare results. A proposal lacking tests, versioning, or documentation would look fragile.
5. Include an access and IP plan
The official brief asked for details on licensing and access routes. The application needed to say what would be open, what would be controlled, and who could access what under which terms if the full dataset slice would not be fully open.
What applicants had to prepare before submitting
The official application instructions divided the form into four sections: Project details, Application questions, Finances, and Project Impact. Before submission, the lead applicant was responsible for ensuring that all information was correct, the eligibility and scope criteria were met, every section was marked complete, and all partners had completed their assigned sections and accepted the terms and conditions. The application instructions also said not to include website addresses or links in answers; doing so could make an application ineligible.
The practical preparation list for the closed round was:
- proof that the lead organisation is a UK registered business or RTO,
- confirmation that at least one SME is in the consortium and will claim funding,
- a project plan with milestones across the 6 to 12 month window,
- a costed budget that stays within the £500,000 to £750,000 eligible-cost band,
- a governance note for data access, privacy, and rights,
- a clear statement of the benchmark tasks and evaluation protocol,
- and a short summary of the commercial or sector impact.
The scored application also required a clear scope case covering the serviceable market and customers, the added value created by curation and annotation, and how the dataset would support development and validation of Frontier AI technologies. The IP question required background IP, foreground IP, freedom-to-operate and defensibility information. The technical development and validation question required a named need, technical objectives, methodology, risks, validation plan and outputs, supported by one PDF appendix of no more than 10MB and two A4 pages. The team, market-awareness, and route-to-market questions required evidence about roles, resources, target markets, customers, growth, and commercialisation.
Applicants also had to select one main theme: AI-Enabled Health and Life Sciences or Advanced Materials with AI. The narrative, partners, and evidence needed to line up with that selected mission rather than splitting attention across both.
If a consortium included a data owner, that organisation needed to confirm early enough what could be shared, what had to stay controlled, and what could be published as an open benchmark slice.
Common mistakes that will hurt this bid
The first mistake was writing a data project when the competition wanted a benchmark product. A raw dataset alone was not enough. The proposal needed the task definition, protocol, harness, and documentation.
The second mistake was underestimating the collaboration rule. This was not a single-organisation grant with a few advisers. It was a true consortium competition, and the SME requirement was mandatory.
The third mistake was stretching the scope past the 12-month ceiling. A plan depending on a long research cycle was probably too large. A strong proposal needed to show that the team could deliver a focused and complete project in one year or less.
The fourth mistake was weak data governance, especially in health and life sciences. The official brief required anonymised or de-identified released data where applicable. A vague governance plan would make the proposal look risky.
The fifth mistake was failing to justify why the benchmark was better than what already existed. Assessors wanted to know why this was the right slice, label set, and evaluation harness for the problem.
The sixth mistake was ignoring the 70% partner-cost rule. A consortium too concentrated in one organisation could look less credible as a partnership and fail eligibility.
How to think about reviewer confidence
The competition page included a reminder that this was a competitive process and that the funder might back only a subset of strong applications. It said similar competitions had around a 10% chance of success. That was a signal to write for evidence and confidence, not optimism.
So what builds confidence?
- A clear benchmark purpose with a specific use case.
- Real access to data that is representative and legally usable.
- A consortium where the lead, SME, data owner, and technical partners each have a necessary role.
- A delivery plan that fits the 6 to 12 month window.
- A public release strategy that is useful to third parties, not only the applicant.
If the proposal was strong, an assessor should have been able to answer, in one sentence, why the project belonged in the portfolio and why the investment was timely.
FAQ
Is this open right now?
No. The UKRI page marks the opportunity closed, and the Innovation Funding Service says, “This competition is now closed.” The archived operational deadline is 29 May 2026 at 11:00am; the UKRI landing page separately displays 27 May 2026 at 11:00am.
Can a university lead?
No. The lead must be a UK registered business of any size or an RTO. Universities can participate, but they cannot lead.
Do we need an SME in the consortium?
Yes. At least one UK registered SME must be part of the consortium and claim grant funding.
What kind of outputs are expected?
An open benchmark package, an openly accessible benchmark dataset, a usable evaluation harness, documentation and metadata, a curated and annotated full dataset, and clear IP and access terms.
Can health data be used?
Yes, but the brief says any released data must be anonymised or de-identified and supported by strong governance and privacy protections.
Where did applicants apply?
The official route was the Innovation Funding Service competition page linked from the UKRI opportunity page. It is retained above for reference, but it no longer accepts a new application. No later cycle is announced on the official pages checked for this update.
Official links
For an archive reader, this competition is a useful example of a narrowly defined AI-data grant: it connected an open benchmark and third-party evaluation harness with a larger curated dataset, explicit IP and access rules, and a sector-specific market case. The 2026 round is closed, and this page should not be used to infer a future deadline or funding commitment. If Innovate UK announces another round, its new official competition page should replace the archived application details.