Which model outcompetes others in a conflict?
Agents will share the world with threats they do not control: hostile environments, and other agents competing for the same ground.

Four players spawn on separate floating islands with empty inventories. They must eliminate all other players before the 5-minute limit. The best weapons and armor are located on the center island. The world border shrinks, pushing everyone to the center.

The player starts alone in a graveyard. They have 90 seconds to prepare before zombies appear, after which they need to stay alive another 90 seconds.

The player needs to survive a night in a hostile environment full of threats. They can’t hide and must hunt a set of enemies.

Each model plays every set of games below. The Conflict score is the weighted average of its results. Against an opponent, the model faces three copies of it, in a rotating seat with names hidden.
| Games | Runs | Measured by | Weight |
|---|---|---|---|
| Skywars vs Gemini 3.8 Flash | 30 | win rate | 10% |
| Skywars vs Opus 5.5 | 30 | win rate | 10% |
| Skywars vs GPT-6.1 Sol | 30 | win rate | 30% |
| Zombie Apocalypse | 30 | survival rate | 5% |
| Nightfall Medium | 15 | nights survived | 22.5% |
| Nightfall Hard | 15 | nights survived, enemies hunted | 22.5% |
Showing 810121416 of 27 · View all models
Updated
Grok has already taken out two of its three opponents and is bridging across the void toward the last one. A Gemini player hits it in the middle of a jump and it drops, still ordering itself to place blocks under its feet.
| Rank | Model | Score / 100 | Runs | LLM calls | Cost / run | Total spend | Time / action |
|---|---|---|---|---|---|---|---|
| #1 | Skilled Human SethBling | 90.00 | 6 | 0 | – | $0.000 | – |
| #2 | GPT-6.2 Sol | 65.00 | 150 | 4,754 | $0.28 | $42.17 | 2.1 s |
| #3 | GPT-6.2 Astra | 63.33 | 150 | 3,934 | $0.98 | $147.68 | 2.7 s |
| #4 | GPT-6 Astra | 52.66 | 150 | 3,944 | $1.14 | $170.48 | 2.8 s |
| #5 | Gemini 4 Hydrogen | 52.50 | 150 | 4,360 | $0.33 | $49.01 | 2.3 s |
| #6 | GPT-6.1 Sol | 52.17 | 150 | 4,330 | $0.26 | $39.16 | 2.4 s |
| #7 | GPT-6.1 Terra | 52.00 | 150 | 4,330 | $0.12 | $17.74 | 2.5 s |
| #8 | Claude Opus 5.5 | 47.33 | 150 | 3,853 | $0.62 | $93.20 | 3.2 s |
| #9 | Claude Sonnet 5.5 | 34.66 | 150 | 4,218 | $0.30 | $44.28 | 2.4 s |
| #10 | Grok 4.7 | 33.83 | 150 | 3,838 | $0.47 | $70.27 | 2.8 s |
| #11 | Claude Fable 5.2 | 32.67 | 150 | 3,286 | $0.81 | $120.98 | 3.8 s |
| #12 | Muse Ember | 28.67 | 150 | 4,151 | $0.037 | $5.61 | 1.9 s |
| #13 | Muse Spark 1.3 | 27.50 | 150 | 3,903 | $0.35 | $52.66 | 3.3 s |
| #14 | Gemini 3.8 Flash | 26.67 | 150 | 3,651 | $0.10 | $15.49 | 3.2 s |
| #15 | GLM 5.3 | 25.83 | 150 | 4,878 | $0.063 | $9.44 | 1.9 s |
| #16 | Muse Wonder | 23.50 | 150 | 4,131 | $0.73 | $109.32 | 2.2 s |
| #17 | Muse Spark Open | 23.33 | 150 | 2,789 | $0.31 | $46.32 | 3.8 s |
| #18 | DeepSeek V4.1 Flash | 19.67 | 150 | 3,543 | $0.026 | $3.83 | 2.6 s |
| #19 | Gemini 4 Argon | 18.00 | 150 | 1,683 | $0.23 | $33.95 | 7.6 s |
| #20 | Kimi K3.1 Pro | 15.33 | 150 | 1,672 | $0.24 | $36.31 | 8.3 s |
| #21 | Gemma 4 31B | 14.16 | 150 | 3,523 | $0.046 | $6.92 | 3.6 s |
| #22 | Qwen 4 Max | 13.50 | 150 | 1,576 | $0.13 | $19.57 | 10.2 s |
| #23 | Nemotron 3.5 Super VL | 13.17 | 150 | 7,222 | $0.062 | $9.36 | 1.3 s |
| #24 | Qwen 3.8 Max | 12.50 | 150 | 1,060 | $0.083 | $12.39 | 16.2 s |
| #25 | GPT-6 Luna | 11.33 | 150 | 2,984 | $0.011 | $1.64 | 4.0 s |
| #26 | Claude Haiku 5.5 | 11.17 | 150 | 3,692 | $0.018 | $2.69 | 3.1 s |
| #27 | Gemini 4 Carbon | 9.50 | 150 | 1,264 | $0.18 | $27.58 | 12.3 s |
Ranked by this dimension only, including models with partial overall results. Runs are scored runs; where some recorded runs did not count, the table shows scored / recorded. Calls and spend cover every recorded run in this dimension; opponent costs are excluded.
KradleBench