Computer Levels 1 to 11: Every Bot, and Exactly How It Is Built
Most sites give you a difficulty slider labelled 1 to 10 and never say what it does. Here is the whole ladder, level by level, with the actual engine settings behind each opponent. Levels 1 to 8 are free; levels 9 to 11 are Pro.
| Level | Bot | Rating | How it plays |
|---|
| 1 | Wobble | 400 | Skill 0, depth 1, 60% random moves |
| 2 | Blinky | 800 | Skill 0, depth 2, 35% random moves |
| 3 | Gizmo | 1100 | Skill 1, depth 3, 15% random moves |
| 4 | Sparky | 1400 | UCI_Elo 1400, depth 10 |
| 5 | Bolt | 1700 | UCI_Elo 1700, depth 10 |
| 6 | Nova | 2000 | UCI_Elo 2000, depth 12 |
| 7 | Phantom | 2300 | UCI_Elo 2300, depth 12 |
| 8 | Titan | 2600 | UCI_Elo 2600, depth 14 |
| 9 | MagnusPRO | 2900 | UCI_Elo 2900, depth 15 |
| 10 | DemonPRO | 3190 | UCI_Elo 3190, depth 16 — the engine's ceiling |
| 11 | SalazarPRO | 3650 | No limit at all, depth 20 |
Looking for level 10?
Here, level 10 is Demon at 3190 — the highest rating Stockfish's strength limiter can be set to. Other sites number their levels differently, so "level 10" is not one opponent: Stockfish's own Skill Level 10 plays at roughly 2786 (see every Stockfish level's rating), and Lichess's levels stop at 8. Pick by rating rather than by level number and you will get the opponent you expect.
Why Weak Computer Opponents Are Hard to Build
Notice that the ladder changes method at 1400. That is not a design flourish — it is forced.
Stockfish has a setting called UCI_Elo for playing at a target rating, and it is genuinely well calibrated. It also has a hard floor of 1320. Ask for 600, 800, 1000 or 1200 and you get the identical opponent every time, because every value below the floor clamps to the same place. Its Skill Level option is no escape either, since the engine ignores it whenever strength limiting is switched on. The full level-to-rating table has the exact figures.
So any site offering a 400-rated opponent has to build it themselves, and the usual shortcut is to make a strong engine search badly. That produces a very particular kind of opponent: it plays twenty accurate moves and then hangs its queen for no reason. It loses, but it never feels like a person losing.
Our three bots below 1320 take a different route. The engine still searches properly, and what changes is the choice: Wobble plays a random legal move 60% of the time, Blinky 35%, Gizmo 15%, each on a deliberately shallow search. A beginner's mistakes come from not looking far enough and not noticing what is there, and that is what shallow depth plus genuine randomness reproduces — not brilliance punctuated by catastrophe.
Which Bot Should You Play?
Pick one that beats you roughly a third of the time. An opponent you always beat teaches you nothing, and one that always beats you teaches you nothing either — you need games where a single decision changes the result.
- New to chess: Wobble (400). It blunders often and lets you practise finishing a won game, which is harder than it sounds.
- Know the rules, lose to tactics: Blinky (800) then Gizmo (1100).
- Play online casually: Sparky (1400) or Bolt (1700) — around the strength of a typical club opponent.
- Rated player: Nova (2000) up to Titan (2600), all calibrated with real
UCI_Elo values. - Want to be crushed: Salazar, full-strength Stockfish 19, no limiting of any kind. You will not win.
Not sure where you sit? The ELO test estimates your rating with the FIDE formula, and then you pick the bot about 200 points above it.
Losing to the Computer Is Only Useful If You Look at Why
Playing a bot over and over does surprisingly little on its own. The games that improve you are the ones you go back through, and against an engine you have a luxury you never get against a person: the opponent can tell you exactly where the game turned.
After a loss, put the position where it went wrong into the calculator and see what you missed, or run the whole game through the analyser, which labels every move and marks the blunders. One game examined properly is worth ten played on autopilot.
