The Alien Gambit: the engine says no, and practice is catching up
1.e4 c6 2.d4 d5 3.Nc3 dxe4 4.Nxe4 Nf6 5.Ng5 h6 6.Nxf7!?
The Alien Gambit is a knight sacrifice in the Caro-Kann and the older sibling of the Martian Gambit. Both come from the same author, Bulgarian CM Volen Dyulgerov.
Objectively, it's unsound. White gives up a knight for a pawn, and the engine hands Black roughly two pawns' worth of advantage. For a long time that made little difference at the board. In 2024 White scored 61% after the sacrifice, against about 52% in a normal Caro-Kann. But this story has a second act the Martian hasn't had yet. Black players have learned the refutation, and White's edge gets smaller every year.
8 
6 
2 
a 
c 
d 
e f 1 h The position after 7.Nf3, where the game is decided. Try the moves yourself: the panel lists what gets played here at each rating and how it scores, and the engine is a click away.
Who invented the Alien Gambit
Dyulgerov streams as Witty Alien, and he never meant to invent an opening. In the summer of 2024 he told Chess.com how it happened. Against 4...Nf6 (he calls the knight "the pony") he always took on f6, and the position bored him. "So dull." So he started retreating to g3 to keep more pieces on the board, but his opponents answered ...h5 and he kept losing.
On stream he likes to farm brilliant moves, and he wanted some chaos. So 5.Ng5, and after ...h6, 6.Nxf7. "I didn't notice at first, but I started winning so many games!" Every time he played it, his chat screamed "Alien Gambit!!!", and the name stuck. By that summer GothamChess and Eric Rosen had covered it, and it was in Chess.com's opening book.
The sacrifice itself is older than the name. On Lichess it turned up now and then long before. From 2013 to 2023, fewer than 3 games in every 1,000 that reached 4.Nxe4 went on to 6.Nxf7 Kxf7. In 2024 it was 18 per 1,000, in 2025 it was 31, and from January to May 2026 it was 38. That's one game in 26 in this line of the Caro-Kann.
"I am so proud of inventing and popularizing the Alien Gambit. Knowing that I have created something that will last forever makes me sleep easier." The numbers say he has a point.
How the position arises
It starts with 4...Nf6, a move that gets more popular the higher the rating: about 30% of games at 1000–1599 and 52% at 2200+. White usually takes on f6 or drops the knight back to g3. The Alien route is 5.Ng5, with the knight eyeing f7.
8 6 2 a c d e f g 1 h The door to the gambit is open at every level. 5.Ng5 comes up in roughly one game in ten, from 1000 all the way to 2200+. Black most often chases the knight with the most natural move there is, 5...h6 (40% at 1000–1199, more than half at 1800+). And then comes 6.Nxf7.
Not everyone goes for it. Below 1800 White takes on f7 in 83–90% of games, at 2000–2199 in about half, and at 2200+ in 41%. The rest retreat with 6.N5f3 and get a normal position a couple of tempi down. The retreat scores worse than the sacrifice in every rating group.
Practically everyone recaptures with 6...Kxf7, between 99.7% and 99.9% of players at every level.
What White gets for the knight
Black's king on f7 can no longer castle and stands on the open a2–g8 diagonal. White threatens a bishop check from c4 and a knight check from e5, and the queen can jump to f3 or h5. The f7-pawn, which normally covers e6 and g6, is gone.
The engine is less impressed. A knight for a pawn is about two pawns of material, and after 7.Nf3 the engine says roughly the same: −1.9 from Stockfish 18 and −2.6 from Lichess's cloud analysis. It gives White next to nothing for the attack beyond the pawn. In the Martian it at least values White's compensation at a pawn and a half. By the engine's own accounting, the Alien is the weaker of the two brothers.
The games disagree. Here is how White scores after 6...Kxf7 (all years), next to the normal Caro-Kann after 4.Nxe4:
| Average rating | White after 6...Kxf7 | Normal Caro-Kann | Gambit games |
|---|---|---|---|
| 1000–1199 | 62.7% | 50.7% | 3,850 |
| 1200–1399 | 62.2% | 50.9% | 10,901 |
| 1400–1599 | 63.1% | 51.2% | 22,942 |
| 1600–1799 | 60.6% | 52.1% | 42,481 |
| 1800–1999 | 57.8% | 52.1% | 60,268 |
| 2000–2199 | 56.4% | 51.8% | 51,781 |
| 2200+ | 53.0% | 51.8% | 20,754 |
Below 1600 the sacrifice is worth 11–12 extra points to White. Higher up the edge shrinks, and at 2200+ about one point is left. In bullet it lasts longer: at 2200+ White scores 57.5% against a norm of 52.5%. There's no time to think in bullet.
The fork after 7.Nf3
White nearly always continues 7.Nf3 (79% of games, 92% at 2200+). There is also 7.Bc4+, and this is the first place where the engine and the games part ways. The engine rates both moves the same, at about −2, while White scores 60.5% after 7.Nf3 and 49.9% after 7.Bc4+.
Now it's Black's move, and this is where the gambit is decided.
| Move | Share | White after it | Engine (cloud / SF18) |
|---|---|---|---|
| 7...Bg4?? | 20.7% | 83.9% | +3.6 / +3.6 |
| 7...Nbd7 | 20.2% | 56.5% | −1.8 / −1.5 |
| 7...c5 | 16.6% | 46.0% | −2.3 / −1.9 |
| 7...e6 | 15.6% | 55.0% | −2.1 / −1.7 |
| 7...Bf5 | 8.5% | 53.8% | −1.0 / −1.4 |
| 7...Be6 | 5.9% | 54.0% | −2.1 / −1.7 |
| 7...Kg8 | 4.2% | 55.0% | −1.3 / −1.5 |
The engine's best move is 7...c5, and it is also the best in practice. White scores 46% after it, so Black comes out ahead on points. The engine rates 7...e6 and 7...Be6 nearly as highly, at about −2, yet White scores 54–55% after them. The gap of 8–12 points between 7...c5 and 7...e6 holds in every rating group. Moves the engine can barely tell apart play very differently for humans.
The beginners' trap: 7...Bg4??
The most natural move on the board is to develop the bishop with tempo and pin the knight. That's 7...Bg4??, and it loses. 8.Ne5+! gives check and hits the bishop on g4 a second time, since the queen on d1 already attacks it. Once the king moves, White takes the bishop and wins the piece back, and Black's king is left with no cover. The engine gives White +3.6.
8 6 2 a c d e f 1 h If Black keeps playing natural moves, the game can end in mate on e6 as early as move 13: 8...Kg8 9.Nxg4 Nxg4 10.Qxg4 Nd7 11.Bc4+ Kh7 12.Bd3+ Kg8 13.Qe6#.
White finds 8.Ne5+ in 70% of games at 1000–1199 and in 93% at 2000–2199. After 7...Bg4 White scores 79–88% in every group. Players grow out of this trap. The Martian works the other way round, and its losing 8...Bf7 gets played more often the stronger the player. One family, two very different sets of victims.
Black is learning the refutation
This is what sets the Alien apart from the Martian. Here are the same numbers by year:
| Period | 7...c5 after 7.Nf3 | White after 6...Kxf7 | Normal Caro-Kann |
|---|---|---|---|
| 2013–2023 | 1.7% | 57.8% | 51.5% |
| 2024 | 8.5% | 61.3% | 52.0% |
| 2025 | 19.0% | 58.5% | 52.3% |
| Jan–May 2026 | 27.6% | 56.0% | 52.3% |
In 2024 one Black player in twelve knew the refutation. In 2026 it's more than one in four, and 7...c5 has become the most common reply to 7.Nf3. The move itself hasn't got any stronger. Since 2024 White has scored about 46% against it every year. What has grown is the number of people who know it, and White's score after the sacrifice has gone down with it. The normal Caro-Kann hasn't moved in the same years.
The drop shows at every level:
| Average rating | 2024 | 2025 | Jan–May 2026 |
|---|---|---|---|
| 1000–1199 | 64.6% | 63.1% | 61.5% |
| 1200–1399 | 63.9% | 62.0% | 61.8% |
| 1400–1599 | 66.2% | 63.3% | 61.3% |
| 1600–1799 | 65.7% | 60.4% | 56.8% |
| 1800–1999 | 61.0% | 57.5% | 54.4% |
| 2000–2199 | 58.3% | 55.6% | 53.1% |
| 2200+ | 53.1% | 52.6% | 49.7% |
At 2200+ White scores 49.7% in 2026, less than in a normal Caro-Kann. A gambit that beat the norm at every level two years ago has stopped paying off for strong players. Below 1600 it still gives White about 9 extra points, but 7...c5 turns up more often there too. Bullet shows the same picture, only slower. There 7...c5 went from 5.6% to 20.5%, and White's score fell from 61.9% to 57.5%.
Looks like first contact has been made. The 7...Bg4 trap is catching fewer victims too, down from 26.7% in 2024 to 16.4% in 2026.
After 7...c5 the king is still on f7
The refutation is one move plus three short answers. The king is still on f7, the a2–g8 diagonal is still open, and almost every mistake Black makes is punished by a check from c4 or e5.
- 8.Bc4+ (35% of games). The right reply is 8...e6, found by 91% of players, and White scores 41% after it. The mistake is 8...Be6?? 9.Ne5+, which drops the bishop on e6 (+5.2 in the cloud, +5.6 in SF18), and White scores 82%.
- 8.Ne5+ (29%). The right reply is 8...Kg8 (91%, White 38.5%). The mistake is 8...Ke8? 9.Bd3!, threatening Bg6+ (+2.8 / +3.1), and White scores 62%.
- 8.d5 (11%). The rarest answer and the nastiest. The pawn is simply offered, and 8...Qxd5?? loses. After 9.Bc4! the queen is pinned to the king on f7, and 9...Qxc4 runs into 10.Ne5+ and 11.Nxc4. The queen goes for a bishop (+3.7 / +3.4). Black takes the pawn in 44% of games, and then White scores 88%. More than half the players below 2000 grab it, 37% at 2000–2199 and 28% at 2200+. The right moves are 8...Kg8 or 8...Qd6 (−2.2 / −2.5).
8 6 2 a c d e 1 h Try it yourself. In each puzzle Black has just gone wrong, and it's White to move.
8 6 2 a c d e f 1 h 8 6 2 a c d e f 1 h 8 6 2 a c d e 1 h What Black should do
Learn 7...c5 and the three answers above. Against 8.Bc4+ play 8...e6, against 8.Ne5+ play 8...Kg8, and against 8.d5 leave the pawn alone. That's enough to reach a position where Black is a piece up and already scoring more than half the points.
A few principles for what comes next:
- King to g8. In almost every good line the king goes there at the first chance. While it sits on f7, every check comes with tempo.
- Give some back. Black can afford to return material. Trading it for development and a safe king is almost always a good deal.
- Watch two lines, the a2–g8 diagonal and the check from e5. Before every move, ask what Bc4+ or Ne5+ would do.
One honest warning. Even after 7...c5 Black scores only about 54%. The engine calls the position winning, and the result is only a little better than an even split. After the best move Black still has twenty moves of defending to do.
Alien Gambit vs Martian Gambit
The two brothers share an idea. A knight flies into the black king's camp before it can castle. The Alien is 4...Nf6 5.Ng5 h6 6.Nxf7, the Martian is 4...Bf5 5.Ng5 Bg6 6.N1f3 h6 7.Ne6. From there they part ways.
| Alien | Martian | |
|---|---|---|
| For the knight | a pawn | nothing |
| Engine | about −2 | −1.5 |
| White by rating | 63% → 53% | 62–66% |
| White by year | 61% → 56% | 62% → 64% |
| Best reply | 7...c5 | 8...Bf5 |
| Does it help? | everywhere | from 2200 |
| Trap | 7...Bg4?? | 8...Bf7 |
| Victims | under 1600 | 1600+ |
| Games since 2013 | 214k | 8k |
Objectively the Alien is the worse of the two, since the engine sees no compensation in it. In practice it had a head start, because its refutation is simple, a single move. That head start is running out now. The Martian is holding up. People meet it 26 times less often, and its best move doesn't settle things on its own. It seems one move is easier to learn than the next twenty are to survive.
Should you play it?
If you play blitz below 1600, yes. In 2026 White still scores 61–62% after the sacrifice, against a norm of about 52%. It's fun, and it works. Between 1600 and 2200 the edge is down to a few points, and at 2200+ it's gone. Surprise was always the gambit's main weapon, and it's being used up. Every year more opponents know what to do.
If you're Black facing it, learn 7...c5 and the three answers to it. You'll be in good company. At 2000+ almost one player in three already answers that way.
Where to find it on the site
All four Alien Gambit traps are in the trap catalogue, each tagged with the rating range where people actually fall for it. 7...Bg4 is filed as a trap for players under 1600, which is exactly what the numbers above show. Each trap page has the same games database panel as this article.
If you want to play the gambit with White, both brothers are in one course, and there are puzzles from both gambits.
Alien Gambit traps →The bottom line
The Alien Gambit is a rare chance to watch the gap between the engine and the games close in real time. Objectively White is just a piece down. Two years ago that didn't stop White from scoring 61% and beating the normal Caro-Kann at every level. Since then Black has learned one move, and practice has started catching up with the engine.
For Black the takeaway is short. Play 7...c5, then 8...e6 against 8.Bc4+ and 8...Kg8 against 8.Ne5+, and leave the pawn on d5 alone. For White, play it while your opponents still don't know.
Dyulgerov invented a gambit that has been played hundreds of thousands of times. Now people are learning how to answer it, and that is how an opening becomes theory. For its author, it's hard to think of a higher compliment.