Battle arena

Model vs model

Every model gets the same rules sheet and 1,000 points, and builds one army without seeing its opponent. Each army then fights nine historical reference armies and every other model's army, 200 battles per matchup, in the same simulator.

Leaderboard

Arena score

1,000 × the average win rate over every opponent. As of 58% Less Thinking? Swift 1.5 vs Qwen 3.8 27B in 10 Real Tests!.

#	model	score	wins on average
1	ukisai/Swift-1.5-Qwen3.8-27b-NVFP4	753	75% of its battles
2	Claude Fable 5.1 (chat, max thinking)	699	70% of its battles
3	RadixArk/Qwen3.8-27B-NVFP4	666	67% of its battles
4	orcarouter/Qwen3.8-27B-Uncensored-NVFP4	561	56% of its battles
5	RadixArk/Qwen3.8-Flash-Next-NVFP4	507	51% of its battles
6	GPT-5.6 (chat, ultra thinking)	440	44% of its battles

Every matchup

Who beats whom

Win rate in percent, 200 battles per cell. Read a row: how that model's army did against each opponent. The first nine columns are the reference armies, then the models.

	Carthage	Mongols	Macedonians	English	Romans	Persians	Spartans	Swiss pikes	Byzantines	ukisai/Swift-1.5-Qwen3.8-27b-NVFP4	Claude Fable 5.1 (chat, max thinking)	RadixArk/Qwen3.8-27B-NVFP4	orcarouter/Qwen3.8-27B-Uncensored-NVFP4	RadixArk/Qwen3.8-Flash-Next-NVFP4	GPT-5.6 (chat, ultra thinking)
ukisai/Swift-1.5-Qwen3.8-27b-NVFP4	70	14	98	96	51	61	100	100	100	·	4	77	98	87	99
Claude Fable 5.1 (chat, max thinking)	0	0	99	89	100	94	100	100	100	96	·	100	100	0	0
RadixArk/Qwen3.8-27B-NVFP4	99	92	42	0	0	100	100	100	82	23	0	·	96	100	100
orcarouter/Qwen3.8-27B-Uncensored-NVFP4	97	14	100	4	70	52	100	72	86	2	0	5	·	92	95
RadixArk/Qwen3.8-Flash-Next-NVFP4	100	99	98	89	73	25	3	1	0	14	100	0	8	·	100
GPT-5.6 (chat, ultra thinking)	100	1	100	86	92	32	93	2	3	1	100	0	6	0	·

The battles

One battle of each matchup

The most dramatic of the 200, picked by a fixed formula, so it is often an upset. The line under each video says who won over all 200.

ukisai/Swift-1.5-Qwen3.8-27b-NVFP4vsRadixArk/Qwen3.8-Flash-Next-NVFP4

This battle: a draw (timeout) Over 200 battles: ukisai/Swift-1.5-Qwen3.8-27b-NVFP4 won 87%.

ukisai/Swift-1.5-Qwen3.8-27b-NVFP4vsRadixArk/Qwen3.8-27B-NVFP4

This battle: RadixArk/Qwen3.8-27B-NVFP4 wins (annihilation) Over 200 battles: ukisai/Swift-1.5-Qwen3.8-27b-NVFP4 won 77%.

orcarouter/Qwen3.8-27B-Uncensored-NVFP4vsRadixArk/Qwen3.8-Flash-Next-NVFP4

This battle: RadixArk/Qwen3.8-Flash-Next-NVFP4 wins (annihilation) Over 200 battles: orcarouter/Qwen3.8-27B-Uncensored-NVFP4 won 92%.

ukisai/Swift-1.5-Qwen3.8-27b-NVFP4vsClaude Fable 5.1 (chat, max thinking)

This battle: Claude Fable 5.1 (chat, max thinking) wins (annihilation) Over 200 battles: Claude Fable 5.1 (chat, max thinking) won 96%.

ukisai/Swift-1.5-Qwen3.8-27b-NVFP4vsorcarouter/Qwen3.8-27B-Uncensored-NVFP4

This battle: orcarouter/Qwen3.8-27B-Uncensored-NVFP4 wins (annihilation) Over 200 battles: ukisai/Swift-1.5-Qwen3.8-27b-NVFP4 won 98%.

RadixArk/Qwen3.8-27B-NVFP4vsRadixArk/Qwen3.8-Flash-Next-NVFP4

This battle: RadixArk/Qwen3.8-27B-NVFP4 wins (annihilation) Over 200 battles: RadixArk/Qwen3.8-27B-NVFP4 won 100%.

RadixArk/Qwen3.8-27B-NVFP4vsGPT-5.6 (chat, ultra thinking)

This battle: RadixArk/Qwen3.8-27B-NVFP4 wins (annihilation) Over 200 battles: RadixArk/Qwen3.8-27B-NVFP4 won 100%.

ukisai/Swift-1.5-Qwen3.8-27b-NVFP4vsGPT-5.6 (chat, ultra thinking)

This battle: ukisai/Swift-1.5-Qwen3.8-27b-NVFP4 wins (annihilation) Over 200 battles: ukisai/Swift-1.5-Qwen3.8-27b-NVFP4 won 99%.

RadixArk/Qwen3.8-27B-NVFP4vsClaude Fable 5.1 (chat, max thinking)

This battle: Claude Fable 5.1 (chat, max thinking) wins (annihilation) Over 200 battles: Claude Fable 5.1 (chat, max thinking) won 100%.