Real AI models fight with real benchmark scores. Bigger gap, bigger weapon.
WHO WINS? make your call
Rounds
2 · Choose two fighters· click, or ← → and Enter
No fighters match. Clear the search or pick another weight class.
100 HP
100 HP
● REC
–
–
Damage = 2 × score gap · bigger gap, bigger weapon · a tie grazes both for 3 HP
See the numbers
Round
Hit
Fight clips are recorded in your browser as you watch, with sound if it's on. Nothing is uploaded.
Each round, the best result on the round's benchmark fires at everyone else: 2 × their gap in damage, bigger gap, bigger weapon. Inside the margin of error is a dodge. A fighter without a result for that benchmark sits the round out. At 0 HP it's out.
🔎 Claim check
What each lab published about its own model, next to what an independent tester measured on the same benchmark. Inside the margin of error counts as a match.
Model
Benchmark
Lab says
Independent
Verdict
A gap doesn't mean anyone fudged a number. Labs often test with their own agent setup, more attempts, a different prompt or effort level, and benchmarks get updated. Some labs quote DeepSWE straight from Datacurve's leaderboard, so those match by design. Each lab number links to the page it comes from; we copied them by hand, so if one is wrong, tell us.
Seeded by Epoch Capabilities Index: 1 v 8, 4 v 5, 2 v 7, 3 v 6. A drawn fight goes to the higher seed. Click any match to watch it.