NATIONS ROYALE BENCHMARK · LOCAL, THINKING LOW

gpt-oss 20B

Openai · open weights · released 2025-08-05 · route gpt-oss:20b · Model profile · Chess results · Sudoku results · Mines results · FreeCell results · Hustle Sim results · Elemental Armory results · Micro Mouse results

Results

Score
10095%: 69 to 100 · 3 of 12 base wars so far
Wars won
3 of 3
Won / lost / timed out
3 / 0 / 0
Share of the world held at the end, on average
81.3%
Decisions per war
87.3
Orders sent per decision
5.80
Orders the game refused
462 of 1,519 (30.4%)
Refused orders per decision
1.76
Decisions with no readable orders
0.0%
Retries for a reply with no orders tag
0
Model minutes per war
875.2
Tokens per decision (in / out)
2,696 / 7,360
Reasoning tokens per decision
n/a
Game build
0.91.0+e426bd2
API-equivalent cost per war
n/a
Seconds per decision
601.3
Replies cut off at the length limit
0.0%
Aborted on our side (not counted)
16

By tier

TierWarsScoreWon / lost / timed outHeldMedian winReference player
Easy11001 / 0 / 079%4:33100
Medium11001 / 0 / 070%8:260
Hard11001 / 0 / 094%8:400
Extreme0n/a0 / 0 / 0n/an/an/a

Against the game's players, war by war

Its war score minus each player's on the same wars, in points out of 100. Better and worse count the wars where it scored higher or lower.

PlayerWarsGapBetter / worse
Reference player3+66.72 / 0
Reference on the model's clock3+66.72 / 0
Sharp persona3+33.31 / 0
Median persona3+65.82 / 0
Tourist persona3+98.33 / 0

Every war