In 2024, the last funded proposal in an average single-stage Horizon Europe call scored 13.51 out of 15. By 2026 it was 14.03 - the highest level on record. Every figure in this analysis comes from published call results across 2,367 research and innovation topics, call years 2023 to 2027, as of October 2026. Nothing here is modelled or estimated.
The short version: applying got cheaper. Winning did not get easier.
- 14.03 / 15 - the 2026 funding bar in single-stage calls
- 52% - share of settled 2026 topics cutting at 14.5 or higher
- +60% - proposals per topic since 2024
- 7.8% - overall success rate in 2026
The six questions
A funding bar can only move for a handful of reasons. Each one leaves a different mark in the published results, so they can be told apart rather than argued about.
| Question | Answer | Evidence |
|---|---|---|
| Is the funding bar actually rising? | Yes | 13.51 to 14.03, holds under every control tested |
| Are evaluation scores inflating? | No | Top band static across four independent checks |
| Are more proposals arriving? | Yes | Up 60% per topic |
| Is funding being cut? | No | Grants flat, budgets delivered in full |
| What connects these to the bar? | Crowding | Explains roughly 70% of the movement |
| Do two-stage calls behave the same? | No | A separate regime, far less crowded |
Why most trackers show the opposite
Averages that mix call formats together report a falling bar - roughly 13.5 for three years and dropping in 2026. Practitioners report the opposite. Both cannot be right.
The explanation is a measurement artefact. A two-stage call publishes two results: a first round scored out of 10 that decides who may submit a full proposal, and a final round scored out of 15 that decides funding. Both land in the same public field with nothing to tell them apart. Averaging a 10-point result with a 15-point one is not a rounding error - it is a category mistake.
Timing makes 2026 the worst affected year: that year's two-stage calls have so far contributed seventeen first-round results on the 10-point scale and no final-round results on the 15-point scale, dragging the naive average down by 0.82 points. Separate the two formats and the 2026 bar is the highest on record, not the lowest. Every figure in this analysis keeps them apart.
The average hides the real change
A half-point rise in the mean understates what applicants face. What moved fastest is the share of topics demanding a near-perfect score.
| Topics requiring | 2024 | 2025 | 2026 |
|---|---|---|---|
| 14.0 or better | 47.6% | 58.3% | 67.4% |
| 14.5 or better | 27.6% | 40.3% | 51.7% |
| A perfect 15.0 | 9.7% | 18.0% | 19.1% |
The share of topics where nothing below 14.5 was funded has nearly doubled in two years, and a fifth now fund only proposals scoring a perfect 15.
Three explanations. Only one survives.
Either proposals score better, more proposals arrive, or less money is on offer. Each predicts something different in the published results.
| Explanation | What it would look like | What the results show | Verdict |
|---|---|---|---|
| Proposals score better | More proposals reaching the top band | No change: 19.0% to 19.0% | Rejected |
| More proposals arrive | Rising submissions per topic | Up 60% per topic | Confirmed |
| Less money on offer | Fewer grants than planned | Grants delivered to plan | Rejected |
Across 5,915 individually scored proposals over three years, the score distribution did not move: the 14-to-15 band was 19.0% of scored proposals in 2024, 18.4% in 2025, 19.0% in 2026.
The AI question, answered directly
The data cannot see who used AI. It can see whether proposals got better - which is what the claim requires. They did not, under four independent tests of the score distribution: the top band, the upper half, the whole distribution at once, and top scorers measured against every proposal submitted rather than only those clearing threshold. All four: no shift. Out of every hundred proposals submitted, the number scoring 14 or better was 12.7 in 2024, 12.2 in 2025, and 11.6 in 2026 - drifting slightly down, not up.
One signal looks at first like improvement. The share of submissions clearing the 10-point quality bar rose from 58.7% in 2024 to 65.3% in 2025. But it then fell back to 58.7% in 2026 - exactly where it started. Technology adoption only ever increases, so a rise that fully reverses is not its signature.
What did move in one direction is the share of submissions rejected before evaluation even begins, which went from 3.7% to 5.9%. A field getting better at writing proposals does not produce more inadmissible ones.
That is the signature of cheaper applications, not better ones.
What actually moved the bar
Qualifying proposals per grant rose 46%, from 5.15 to 7.53. Almost the entire increase is volume: more proposals submitted accounts for more than the whole move, slightly offset by the modest rise in grants awarded. The share clearing threshold contributed nothing at all, because it ended the period where it began.
Across individual topics, the more qualifying proposals competing per grant, the higher the cutoff. Once that relationship is accounted for, most of the year-on-year rise disappears - held at 2024 competition levels, the 2026 bar would have been 13.65 instead of 14.03. Roughly 70% of the rise is crowding alone. In money terms: the funding requested by qualifying proposals rose from a median of 47 million euro per topic to 99 million. Same grants, far more demand chasing them.
The full funnel:
| Stage | 2024 | 2025 | 2026 | Change |
|---|---|---|---|---|
| Proposals submitted per topic | 20.77 | 22.16 | 33.23 | +60% |
| Grants awarded per topic | 2.37 | 2.68 | 2.59 | +9% |
| Qualifying proposals per grant | 5.15 | 5.39 | 7.53 | +46% |
| Funded among those qualifying | 19.4% | 18.5% | 13.3% | -31% |
| Overall success rate | 11.4% | 12.1% | 7.8% | -32% |
Where it bites hardest
Success rates differ five-fold between parts of the programme, so where you apply now matters more than it did in 2024.
| Cluster | 2024 | 2025 | 2026 | Direction |
|---|---|---|---|---|
| Cluster 6 Food, Bioeconomy & Environment | 12.9% | 10.1% | 5.8% | Falling |
| Cluster 5 Climate, Energy & Mobility | 12.9% | 14.0% | 7.5% | Falling |
| Cluster 4 Digital, Industry & Space | 10.5% | 7.6% | 7.2% | Falling |
| Joint Undertakings | 11.0% | 21.0% | 19.6% | Rising |
| Missions | 10.6% | 12.0% | - | No clear trend |
| Research Infrastructures | 8.1% | 39.8% | - | Rising |
Clusters 4, 5 and 6 all decline; Joint Undertakings move the other way. And 24% of settled 2026 topics now fund fewer than 1 in 20 of the proposals they attract - near-lotteries, against roughly 5% in each of the two previous years.
The clearest strategic opening: two-stage calls
Two-stage calls are becoming rarer - 14.9% of topics in 2024, 8.0% in 2026 - but where they survive, they are far less crowded. The first round is a filter, not a competition for money: it admits a median 58% of applicants at a bar of 8.8 out of 10. In the final round, 43% of proposals above threshold are funded, against 13 to 19% in single-stage calls. Where a two-stage call exists, reaching the final round is worth far more than the same effort spent on a crowded single-stage topic. (Small samples - 26 first-round and 46 final-round results, directional guidance rather than a precise estimate.)
What this analysis cannot say
The direction of every headline finding is well supported. Precise levels for 2026 are not, and the cause behind the volume rise is not identifiable from this data.
- Three years is a short series; one unusual year moves the estimates.
- 89 of 320 single-stage 2026 topics have settled results so far, concentrated in Clusters 4, 5 and 6. The 2026 figures will keep moving as the rest arrive.
- Two-stage findings rest on 26 first-round and 46 final-round results - directional only.
- Cluster 3 Civil Security has no published results in any year, across 146 topics, so it is absent throughout.
- No proposal texts or individual scores exist in this data, so AI use is never observed - only the traces it would leave.
Evaluators scoring the middle of the field more generously, broader topic definitions, or heavier resubmission would each produce a similar pattern, and none can be separated without applicant-level data. What is settled is narrower and still useful: proposals are not scoring better, funding has not been cut, and there are far more applications competing for each grant than there were in 2024.
Method, briefly
Built from published call results for 2,367 Horizon Europe research and innovation topics. Single-stage and two-stage calls are never averaged together; first-round results never appear in any 15-point figure; ratios are computed only where a topic's own funnel is internally consistent; 2023 is excluded as too small to measure and 2027 has no results yet.
One source-data caveat: for a handful of calls the published figures repeat a call-wide submission total against every topic in that call, rather than the per-topic count. Those rows are excluded, because left in they distort submission and success rates substantially.
The full analysis, including every cleaning rule and the code that produces each number above, is published as a notebook in the GrantSignal repository. Re-running it against current data is how this article is kept accurate.