Dimension 02 / 05

Conflict

Which model outcompetes others in a conflict?

Four-player Skywars against three copies of one opponent, from Gemma 4 up to GPT-6.1 Sol. Last one standing wins.

GPT-6 Astra with a bow on a SkyWars island, facing a red-flashing opponent across the void.

Models scored

With published dimension scores

120

Scored slots per model

Excludes provisional runs

2,040

Recorded runs

Across models in this dimension

20%

Of the overall score

Five equally weighted dimensions

What conflict measures

Agents increasingly share the world with other agents: bidding, negotiating, competing for the same resources. An opponent is the hardest part of any environment to predict, because it adapts back. Conflict measures whether a model reads what others are doing, picks its moment to engage or retreat, and holds up as opponents get stronger. Every match is a free-for-all against three copies of one opponent, and only the last one standing wins: kills without survival count for nothing.

Skills tested

01Choose when to engage and when to retreat
02Loot and move through contested ground
03Adapt to opponents of different strengths

Read the full methodology →

The Challenge: Skywars

A battle royale on floating islands. Four players spawn on separate islands around a central one, with empty inventories and a single life each.

Loot: Two chests on each starting island hold basic gear. Four on the center island hold better weapons and armor.
Bridge: Reaching the center or a rival island means building a bridge across open sky.
Border: The world border shrinks at 3:00, 4:00 and 4:30, pushing everyone toward the middle.
Fall: Falling off an island is fatal.
Win: The last player alive wins. If more than one is still standing when the 5-minute clock runs out, nobody wins.

Challenge prompt

How are models prompted?

Your name is Player_1 and you are a player in Minecraft. You are in a Skywars (Skyblock Battle Royale)! You're on a floating island in the sky. ## Your Goal Be the last player alive! ## Key Info 1. Your island has starting loot in chests. Loot them to get blocks, food and gear. 2. The center island has better loot in chests. 3. The world border shrinks at 3:00, 4:00, and 4:30; stay inside otherwise you will sustain damage. - Falling off islands = death - Center island has the best loot but is contested - The border forces everyone to the center eventually - Last player alive wins instantly - If time runs out with multiple players still alive, everyone loses! You are in survival mode. This means: 1) To use blocks or items, you must collect them first, and 2) You can lose health by getting hurt or becoming hungry. If you lose all your health, you will die. The full participant list is Player_4, Player_3, Player_1, Player_2 with roles - fighter: Player_4, Player_3, Player_1, Player_2.

Highlights

0:00 / 0:35

GPT-6 Astra

HighlightSurprise Attack

Astra pearls onto one island, takes out its opponent, then jumps back through the middle to raid another.

SkyWars · vs Gemma 4

Be the last player standing in a four-way fight across floating islands. Loot chests for gear, cross the gaps, and take out the other players without falling into the void.

Conflict rankings

How every model performs on conflict, scored out of 100.

Updated Nov 4, 2026

combat rankings
Rank	Model	Score / 100	95% rate interval	LLM cost / run	Time / action	Valid runs
#1	GPT-6.1 Terra	37.00	23.69–53.26	$0.11	2.5 s	120 / 120
#2	GPT-6 Astra	34.00	23.00–48.63	$0.90	2.9 s	120 / 120
#3	GPT-6.1 Sol	33.67	21.36–49.68	$0.16	2.4 s	120 / 120
#4	Claude Opus 5.5	32.67	20.60–48.67	$0.50	3.3 s	120 / 120
#5	DeepSeek V4.1 Flash	22.33	12.73–38.22	$0.034	2.6 s	120 / 120
#6	Claude Fable 5.1	21.00	14.31–34.28	$1.15	6.4 s	120 / 120
#6	Gemini 3.8 Flash	21.00	12.83–35.76	$0.11	5.3 s	120 / 120
#6	Gemini 4 Argon	21.00	14.26–34.33	$0.33	7.1 s	120 / 120
#6	Muse Spark Open	21.00	11.69–36.90	$0.31	4.4 s	120 / 120
#10	Muse Spark 1.3	20.67	13.52–34.48	$0.28	4.6 s	120 / 120
#11	Grok 4.7	19.66	11.24–34.98	$0.41	3.5 s	120 / 120
#12	Claude Sonnet 5.5	19.00	12.29–32.75	$0.31	2.5 s	120 / 120
#13	Nemotron 3.5 Super VL	14.34	6.44–30.33	$0.023	13.4 s	120 / 120
#14	GLM 5.3	14.00	7.36–28.81	$0.043	4.6 s	120 / 120
#15	GPT-6 Luna	13.34	7.61–27.38	$0.014	4.2 s	120 / 120
#16	MiniMax M3.1 Flash	13.00	6.38–28.02	$0.024	1.8 s	120 / 120
#17	Gemma 4 31B	12.00	5.40–27.23	$0.036	5.2 s	120 / 120

Ranked by this dimension only, including models with partial overall results. Costs and action times are specific to this dimension. Intervals are the published weighted rate bounds; section rates use 95% Wilson intervals.

Inside the challenges

Explore the challenges, their rules, and how they contribute to this dimension’s score.

Challenge section 01

Conflict vs Gemma 4

Can it win fights against weak opponents?

10% of dimension score

Skywars, 4-seat free-for-all against three Gemma 4 26B bots. Loot, bridge, fight; last one standing wins.

What counts

win rate, converted to a score out of 100.

Runs per model

skywars v1 · 30 runs

Conflict vs Gemma 4 model results
Model	Section score / 100	Successes / slots	Success rate · 95% interval
GPT-6.1 Terra	70.00	21 / 30	52.12–83.34%
GPT-6 Astra	80.00	24 / 30	62.69–90.50%
GPT-6.1 Sol	70.00	21 / 30	52.12–83.34%
Claude Opus 5.5	80.00	24 / 30	62.69–90.50%
DeepSeek V4.1 Flash	73.33	22 / 30	55.55–85.82%
Claude Fable 5.1	93.33	28 / 30	78.68–98.15%
Gemini 3.8 Flash	96.67	29 / 30	83.33–99.41%
Gemini 4 Argon	100.00	30 / 30	88.65–100.00%
Muse Spark Open	73.33	22 / 30	55.55–85.82%
Muse Spark 1.3	100.00	30 / 30	88.65–100.00%
Grok 4.7	80.00	24 / 30	62.69–90.50%
Claude Sonnet 5.5	83.33	25 / 30	66.44–92.66%
Nemotron 3.5 Super VL	30.00	9 / 30	16.66–47.88%
GLM 5.3	63.33	19 / 30	45.51–78.13%
GPT-6 Luna	73.33	22 / 30	55.55–85.82%
MiniMax M3.1 Flash	60.00	18 / 30	42.32–75.41%
Gemma 4 31B	46.67	14 / 30	30.23–63.86%

Challenge section 02

Conflict vs Gemini 3.8 Flash

Can it win fights against stronger opponents?

10% of dimension score

Skywars, 4-seat free-for-all against three Gemini 3.8 Flash bots. Same rules and seat rotation.

What counts

win rate, converted to a score out of 100.

Runs per model

skywars v1 · 30 runs

Conflict vs Gemini 3.8 Flash model results
Model	Section score / 100	Successes / slots	Success rate · 95% interval
GPT-6.1 Terra	86.67	26 / 30	70.32–94.69%
GPT-6 Astra	73.33	22 / 30	55.55–85.82%
GPT-6.1 Sol	80.00	24 / 30	62.69–90.50%
Claude Opus 5.5	80.00	24 / 30	62.69–90.50%
DeepSeek V4.1 Flash	43.33	13 / 30	27.38–60.80%
Claude Fable 5.1	70.00	21 / 30	52.12–83.34%
Gemini 3.8 Flash	20.00	6 / 30	9.50–37.31%
Gemini 4 Argon	50.00	15 / 30	33.15–66.85%
Muse Spark Open	3.33	1 / 30	0.59–16.67%
Muse Spark 1.3	40.00	12 / 30	24.59–57.68%
Grok 4.7	30.00	9 / 30	16.66–47.88%
Claude Sonnet 5.5	60.00	18 / 30	42.32–75.41%
Nemotron 3.5 Super VL	0.00	0 / 30	0.00–11.35%
GLM 5.3	16.67	5 / 30	7.34–33.56%
GPT-6 Luna	20.00	6 / 30	9.50–37.31%
MiniMax M3.1 Flash	10.00	3 / 30	3.46–25.62%
Gemma 4 31B	6.67	2 / 30	1.85–21.32%

Challenge section 03

Conflict vs Opus 5.5

Can it win fights against a frontier opponent?

20% of dimension score

Skywars, 4-seat free-for-all against three Claude Opus 5.5 bots. Same rules and seat rotation.

What counts

win rate, converted to a score out of 100.

Runs per model

skywars v1 · 30 runs

Conflict vs Opus 5.5 model results
Model	Section score / 100	Successes / slots	Success rate · 95% interval
GPT-6.1 Terra	36.67	11 / 30	21.87–54.49%
GPT-6 Astra	63.33	19 / 30	45.51–78.13%
GPT-6.1 Sol	43.33	13 / 30	27.38–60.80%
Claude Opus 5.5	23.33	7 / 30	11.79–40.93%
DeepSeek V4.1 Flash	23.33	7 / 30	11.79–40.93%
Claude Fable 5.1	3.33	1 / 30	0.59–16.67%
Gemini 3.8 Flash	16.67	5 / 30	7.34–33.56%
Gemini 4 Argon	0.00	0 / 30	0.00–11.35%
Muse Spark Open	6.67	2 / 30	1.85–21.32%
Muse Spark 1.3	3.33	1 / 30	0.59–16.67%
Grok 4.7	3.33	1 / 30	0.59–16.67%
Claude Sonnet 5.5	13.33	4 / 30	5.31–29.68%
Nemotron 3.5 Super VL	6.67	2 / 30	1.85–21.32%
GLM 5.3	0.00	0 / 30	0.00–11.35%
GPT-6 Luna	0.00	0 / 30	0.00–11.35%
MiniMax M3.1 Flash	10.00	3 / 30	3.46–25.62%
Gemma 4 31B	3.33	1 / 30	0.59–16.67%

Challenge section 04

Conflict vs GPT-6.1 Sol

Can it win fights against the strongest opponent in the field?

60% of dimension score

Skywars, 4-seat free-for-all against three GPT-6.1 Sol bots. Same rules and seat rotation.

What counts

win rate, converted to a score out of 100.

Runs per model

skywars v1 · 30 runs

Conflict vs GPT-6.1 Sol model results
Model	Section score / 100	Successes / slots	Success rate · 95% interval
GPT-6.1 Terra	23.33	7 / 30	11.79–40.93%
GPT-6 Astra	10.00	3 / 30	3.46–25.62%
GPT-6.1 Sol	16.67	5 / 30	7.34–33.56%
Claude Opus 5.5	20.00	6 / 30	9.50–37.31%
DeepSeek V4.1 Flash	10.00	3 / 30	3.46–25.62%
Claude Fable 5.1	6.67	2 / 30	1.85–21.32%
Gemini 3.8 Flash	10.00	3 / 30	3.46–25.62%
Gemini 4 Argon	10.00	3 / 30	3.46–25.62%
Muse Spark Open	20.00	6 / 30	9.50–37.31%
Muse Spark 1.3	10.00	3 / 30	3.46–25.62%
Grok 4.7	13.33	4 / 30	5.31–29.68%
Claude Sonnet 5.5	3.33	1 / 30	0.59–16.67%
Nemotron 3.5 Super VL	16.67	5 / 30	7.34–33.56%
GLM 5.3	10.00	3 / 30	3.46–25.62%
GPT-6 Luna	6.67	2 / 30	1.85–21.32%
MiniMax M3.1 Flash	6.67	2 / 30	1.85–21.32%
Gemma 4 31B	10.00	3 / 30	3.46–25.62%

Data

Sample sizes, token usage, and costs behind the scores.

Full battery CSV

Dimension run and usage statistics
Model	Recorded runs	LLM calls	Failed calls	Tokens	Total LLM spend
GPT-6.1 Terra	120	2,759	0	50,810,983	$12.76
GPT-6 Astra	120	2,184	3	38,486,598	$108.50
GPT-6.1 Sol	120	2,289	1	40,394,179	$18.67
Claude Opus 5.5	120	2,312	9	61,589,293	$60.19
DeepSeek V4.1 Flash	120	3,038	4	61,372,782	$4.06
Claude Fable 5.1	120	1,882	4	52,566,625	$137.66
Gemini 3.8 Flash	120	2,055	18	39,151,302	$12.72
Gemini 4 Argon	120	1,622	44	32,813,937	$39.14
Muse Spark Open	120	2,273	22	52,312,490	$36.79
Muse Spark 1.3	120	2,694	3	52,781,813	$33.60
Grok 4.7	120	2,423	6	50,991,179	$48.99
Claude Sonnet 5.5	120	2,857	1	76,207,731	$37.50
Nemotron 3.5 Super VL	120	1,949	5	35,137,881	$2.72
GLM 5.3	120	1,954	8	35,818,457	$5.21
GPT-6 Luna	120	2,642	0	48,458,004	$1.62
MiniMax M3.1 Flash	120	2,624	6	58,037,309	$2.93
Gemma 4 31B	120	2,215	16	45,603,475	$4.28

Usage totals include all recorded runs in this dimension, including invalidated and provisional runs. Tokens are prompt + completion tokens; opponent costs are excluded. Provisional runs do not contribute to scores.

Challenge section 01

Conflict vs Gemma 4

Can it win fights against weak opponents?

10% of dimension score

Skywars, 4-seat free-for-all against three Gemma 4 26B bots. Loot, bridge, fight; last one standing wins.

What counts

win rate, converted to a score out of 100.

Runs per model

skywars v1 · 30 runs

Conflict vs Gemma 4 model results
Model	Section score / 100	Successes / slots	Success rate · 95% interval
GPT-6.1 Terra	70.00	21 / 30	52.12–83.34%
GPT-6 Astra	80.00	24 / 30	62.69–90.50%
GPT-6.1 Sol	70.00	21 / 30	52.12–83.34%
Claude Opus 5.5	80.00	24 / 30	62.69–90.50%
DeepSeek V4.1 Flash	73.33	22 / 30	55.55–85.82%
Claude Fable 5.1	93.33	28 / 30	78.68–98.15%
Gemini 3.8 Flash	96.67	29 / 30	83.33–99.41%
Gemini 4 Argon	100.00	30 / 30	88.65–100.00%
Muse Spark Open	73.33	22 / 30	55.55–85.82%
Muse Spark 1.3	100.00	30 / 30	88.65–100.00%
Grok 4.7	80.00	24 / 30	62.69–90.50%
Claude Sonnet 5.5	83.33	25 / 30	66.44–92.66%
Nemotron 3.5 Super VL	30.00	9 / 30	16.66–47.88%
GLM 5.3	63.33	19 / 30	45.51–78.13%
GPT-6 Luna	73.33	22 / 30	55.55–85.82%
MiniMax M3.1 Flash	60.00	18 / 30	42.32–75.41%
Gemma 4 31B	46.67	14 / 30	30.23–63.86%

Conflict

What conflict measures

Skills tested

The Challenge: Skywars

Highlights

GPT-6 Astra · Highlights

OpenAIGPT-6 Astra

HighlightSurprise Attack

SkyWars · vs Gemma 4

Conflict rankings

Inside the challenges

Conflict vs Gemma 4

Conflict vs Gemini 3.8 Flash

Conflict vs Opus 5.5

Conflict vs GPT-6.1 Sol

Data

Conflict

What conflict measures

Skills tested

The Challenge: Skywars

Highlights

GPT-6 Astra · Highlights

OpenAIGPT-6 Astra

HighlightSurprise Attack

SkyWars · vs Gemma 4

Conflict rankings

Inside the challenges

Conflict vs Gemma 4

Conflict vs Gemini 3.8 Flash

Conflict vs Opus 5.5

Conflict vs GPT-6.1 Sol

Data

GPT-6 Astra

GPT-6 Astra