These results were achieved with SoundIntel 1.1, prior to the developments introduced in SoundIntel 2.0. SoundIntel 1.1 analysed only the first 60 seconds of each track, from audio alone, with a reference dataset of fewer than one hundred tracks and a simulated audience modelled for global listening rather than Eurovision voting culture. The aggregated bookmaker market operated on twelve or more parallel signals (Spotify streams, YouTube engagement, fan polls, expert rankings, betting sentiment) with full access to artist history, country of origin, staging rehearsals, and contest context. Reaching 75% under those input constraints, in an environment designed around factors the engine deliberately does not model, is the number SoundIntel reports.
Illustrative signals
Three tracks up close.
✓ Aligned
Bulgaria (Grand Final winner)
DARA, Bangaranga
Pre-contest bookmaker consensus gave Bulgaria roughly 1% to win Eurovision 2026. SoundIntel detected Bulgaria as a Qualify from Semi-Final 2, five days before the Grand Final, and flagged two features relevant to broad commercial reach: broad audience appeal across segments, and low listener fatigue (a track audiences return to across repeat listens). Bulgaria won the contest with 516 points, taking both the jury vote (204) and the televote (312).
✓ Aligned
Australia (Grand Final, 4th)
Delta Goodrem, Eclipse
SoundIntel registered Australia as the field’s strongest signal for audiences that engage deeply with music, the listener type that behaves most like a professional jury. Australia finished fourth, with a particularly strong jury share. The alignment between the engine’s deep-listener signal and the professional jury verdict is the kind of mapping SoundIntel is designed to produce.
✓ Aligned
Belgium (Semi-Final 1)
ESSYLA, Dancing on the Ice
Pre-contest bookmaker consensus gave Belgium 34% to qualify, ranking it 13th of 15. SoundIntel detected Belgium as a Qualify one day before Semi-Final 1, with a particularly strong signal toward younger audience segments. Belgium qualified. The clearest sleeper signal of the semi-final: a track the market had written off, flagged by the audio-only engine.
Grand Final
16 May 2026, Wiener Stadthalle.
The Grand Final took place on 16 May 2026 at the Wiener Stadthalle, Vienna. Bulgaria won with 516 points, winning both the national jury vote (204 points) and the public televote (312 points). First Bulgarian victory.
Top five
| Rank | Country | Total | Jury | Televote |
|---|
| 1 | Bulgaria | 516 | 204 | 312 |
| 2 | Israel | 343 | 123 | 220 |
| 3 | Romania | 296 | 64 | 232 |
| 4 | Australia | 287 | 165 | 122 |
| 5 | Italy | 281 | 134 | 147 |
Official full scoreboard: eurovision.tv.
On final placement: SoundIntel does not predict Grand Final placement. Placement reflects national jury voting, televote geography, diaspora voting blocs, live performance, staging, running order, and campaign effects, all of which are outside audio scope by design.
View full per-track data for both semi-finals
Semi-Final 1
12 May 2026 · 15 competing entries · 10 qualify
| Country · artist and song | Actual outcome |
|---|
| SoundIntel detection: Qualify |
| BelgiumESSYLA, Dancing on the Ice | ✓ Qualified |
| SerbiaLAVINA, Kraj Mene | ✓ Qualified |
| FinlandLinda Lampenius x Pete Parkkonen, Liekinheitin | ✓ Qualified |
| SwedenFELICIA, My System | ✓ Qualified |
| LithuaniaLion Ceccah, Sólo Quiero Más | ✓ Qualified |
| MoldovaSatoshi, Viva, Moldova! | ✓ Qualified |
| PolandALICJA, Pray | ✓ Qualified |
| GreeceAkylas, Ferto | ✓ Qualified |
| San MarinoSENHIT, Superstar | – Did not qualify |
| EstoniaVanilla Ninja, Too Epic To Be True | – Did not qualify |
| SoundIntel detection: Not Qualify |
| PortugalBandidos do Cante, Rosa | – Did not qualify |
| MontenegroTamara Živković, Nova Zora | – Did not qualify |
| GeorgiaBzikebi, On Replay | – Did not qualify |
| IsraelNoam Bettan, Michelle | ✓ Qualified |
| CroatiaLELEK, Andromeda | ✓ Qualified |
SF1 result: 8 of 10 SoundIntel-detected qualifiers made it through to the Grand Final.
Semi-Final 2
14 May 2026 · 15 competing entries · 10 qualify
| Country · artist and song | Actual outcome |
|---|
| SoundIntel detection: Qualify |
| AustraliaDelta Goodrem, Eclipse | ✓ Qualified |
| CzechiaDaniel Žižka, Crossroads | ✓ Qualified |
| BulgariaDARA, Bangaranga | ✓ Qualified |
| CyprusAntigoni, JALLA | ✓ Qualified |
| DenmarkSøren Torpegaard Lund, Før Vi Går Hjem | ✓ Qualified |
| UkraineLELÉKA, Ridnym | ✓ Qualified |
| RomaniaAlexandra Căpitănescu, Choke Me | ✓ Qualified |
| AzerbaijanJIVA, Just Go | – Did not qualify |
| SwitzerlandVeronica Fusaro, Alice | – Did not qualify |
| ArmeniaSIMÓN, Paloma Rumba | – Did not qualify |
| SoundIntel detection: Not Qualify |
| LuxembourgEva Marija, Mother Nature | – Did not qualify |
| LatviaAtvara, Ēnā | – Did not qualify |
| AlbaniaAlis, Nân | ✓ Qualified |
| MaltaAIDAN, Bella | ✓ Qualified |
| NorwayJONAS LOVV, YA YA YA | ✓ Qualified |
SF2 result: 7 of 10 SoundIntel-detected qualifiers made it through to the Grand Final.
Verification: official qualifier lists for both semi-finals are published at eurovision.tv and eurovisionworld.com.
Methodology note
For each track, SoundIntel returns a per-track output. The SoundIntel detection column applies a pre-registered mapping of that output to Qualify or Not Qualify. The mapping was defined before each semi-final and applied uniformly across both. The mapping specification is available in the Technical Validation Report under NDA.
Scope and limitations
- What SoundIntel measures: audio.
- What SoundIntel does not model: Eurovision outcome factors outside audio scope. National jury voting, televote geography, diaspora voting blocs, live performance, staging, running order, campaign effects, contest narrative. These influence Eurovision results significantly, often more than the music itself.
- What SoundIntel was not built for: the engine was not designed to replicate Eurovision scoring, Eurovision taste, or Eurovision voting culture. It was designed to detect audio-driven audience response at a global level. Eurovision was used here because it is a public, time-stamped, verifiable event, not because it is the engine’s native use case.
- Why this is a hard test: Eurovision is deliberately a hostile environment for an audio-only engine. The contest result reflects far more than the audio of each entry. Audio-only detection of 15 of 20 qualifiers under these conditions, with a sub-100-track reference dataset and a globally modelled simulated audience, is the number SoundIntel reports.
Prediction time-stamp: 11 May 2026. Contest dates: 12, 14, 16 May 2026, Wiener Stadthalle, Vienna. Public outcome sources: eurovision.tv, eurovisionworld.com. Bookmaker data source: eurovisionworld.com/odds, retrieved 11 May 2026. The SoundIntel engine and its implementation are proprietary.