Benchmark Almanac

KradleBench

GPT-6.2 Astra 78.3 leads since November 2026
Not in model cards

Models play live multi-agent Minecraft games and act by writing code. The overall score weights collaboration, economy and conflict at 30% each, and logic and creativity at 5% each.

Kradle launched it on November 18, 2026 with 26 models and one skilled human player (battery kb-1.8.2). Runs last 2 to 15 minutes, and every model uses the same harness, prompt and skills.

Made by
Kradle
Released
November 2026
Kind
Multi-agent games
Level
Everyday
Scored
Settled by events
Answers
Agent task
Measures
Capability
Language
English
In and out
Text, Code → Actions, Code
Interaction
Multi-agent
Environment
Minecraft
Items from
Live play or votes
Refresh
Live
Run by
A host
In views
Games, Agents, Kept fresh
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
26
Models scored on its own leaderboard

GPT-6.2 Astra leads with 78.3

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 1 GPT-6.2 Astra 78.3
  2. 2 GPT-6.2 Sol 67.9
  3. 3 GPT-6 Astra 56.1
  4. 4 GPT-6.1 Sol 54.6
  5. 5 GPT-6.1 Terra 48.6
  6. 6 Gemini 4 Hydrogen 43.9
  7. 7 Claude Opus 5.5 39.1
  8. 8 Claude Fable 5.2 37.8
  9. 9 Claude Sonnet 5.5 33.8
  10. 10 Gemini 4 Argon 28.6
  11. 11 Muse Wonder 27.6
  12. 12 Muse Ember 25.7
  13. 13 Gemini 3.8 Flash 24.3
  14. 14 Grok 4.7 23.0
  15. 15 Gemini 4 Carbon 22.0
  16. 16 Muse Spark 1.3 21.7
  17. 17 GLM-5.3 21.2
  18. 18 DeepSeek V4.1 Flash 19.8
  19. 19 GPT-6 Luna 18.1
  20. 20 Gemma 4 31B IT 17.1
  21. 21 Kimi K3.1 Pro 15.5
  22. 22 Muse Spark Open 15.2
  23. 23 Claude Haiku 5.5 15.0
  24. 24 Nemotron 3.5 Super VL 11.2
  25. 25 Qwen4-Max 11.0
  26. 26 Qwen 3.8 Max 8.68

Source: Official KradleBench leaderboard

#ModelScoreSettingDateSource
1 GPT-6.2 Astra 78.3 Nov 2026 Official leaderboard
2 GPT-6.2 Sol 67.9 Nov 2026 Official leaderboard
3 GPT-6 Astra 56.1 Sep 2026 Official leaderboard
4 GPT-6.1 Sol 54.6 Sep 2026 Official leaderboard
5 GPT-6.1 Terra 48.6 Oct 2026 Official leaderboard
6 Gemini 4 Hydrogen 43.9 Nov 2026 Official leaderboard
7 Claude Opus 5.5 39.1 Sep 2026 Official leaderboard
8 Claude Fable 5.2 37.8 Nov 2026 Official leaderboard
9 Claude Sonnet 5.5 33.8 Sep 2026 Official leaderboard
10 Gemini 4 Argon 28.6 Sep 2026 Official leaderboard
11 Muse Wonder 27.6 Oct 2026 Official leaderboard
12 Muse Ember 25.7 Nov 2026 Official leaderboard
13 Gemini 3.8 Flash 24.3 Sep 2026 Official leaderboard
14 Grok 4.7 23.0 Sep 2026 Official leaderboard
15 Gemini 4 Carbon 22.0 Oct 2026 Official leaderboard
16 Muse Spark 1.3 21.7 Sep 2026 Official leaderboard
17 GLM-5.3 21.2 Aug 2026 Official leaderboard
18 DeepSeek V4.1 Flash 19.8 Sep 2026 Official leaderboard
19 GPT-6 Luna 18.1 Sep 2026 Official leaderboard
20 Gemma 4 31B IT 17.1 Apr 2026 Official leaderboard
21 Kimi K3.1 Pro 15.5 Oct 2026 Official leaderboard
22 Muse Spark Open 15.2 Oct 2026 Official leaderboard
23 Claude Haiku 5.5 15.0 Oct 2026 Official leaderboard
24 Nemotron 3.5 Super VL 11.2 Oct 2026 Official leaderboard
25 Qwen4-Max 11.0 Nov 2026 Official leaderboard
26 Qwen 3.8 Max 8.68 Aug 2026 Official leaderboard

Rankings by part

We read 5 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Collaboration, Economy and Conflict. GPT-6.2 Astra leads 4 of them, and GPT-6.2 Sol leads the other one. Arrows show how far a model moved from its overall rank among the models we track.

Collaboration

26 models we track have a score here.

  1. 1 GPT-6.2 Astra 74.3
  2. 2 GPT-6.2 Sol 70.2
  3. 3 GPT-6.1 Sol ↑1 61.7
  4. 4 GPT-6 Astra ↓1 59.7
  5. 5 Claude Fable 5.2 ↑3 56.0
  6. 6 GPT-6.1 Terra ↓1 54.9
  7. 7 Claude Opus 5.5 48.3
  8. 8 Gemini 4 Argon ↑2 47.6
  9. 9 Gemini 4 Hydrogen ↓3 43.8
  10. 10 Claude Sonnet 5.5 ↓1 40.4

The best score, 78.3, costs $1.26 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek OpenAI Meta Cost frontier One model at each effort level
0.00 20.0 40.0 60.0 80.0 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6.2 Astra GPT-6.2 Sol GPT-6.1 Sol GPT-6.1 Terra Muse Ember DeepSeek V4.1 Flash GPT-6 Luna 0.00 20.0 40.0 60.0 80.0 $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6.2 Astra GPT-6.2 Sol

Source: Official KradleBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Astra 78.3 $1.26
GPT-6.2 Sol 67.9 $0.35
GPT-6 Astra 56.1 $1.55
GPT-6.1 Sol 54.6 $0.32
GPT-6.1 Terra 48.6 $0.16
Gemini 4 Hydrogen 43.9 $0.53
Claude Opus 5.5 39.1 $0.95
Claude Fable 5.2 37.8 $1.35
Claude Sonnet 5.5 33.8 $0.43
Gemini 4 Argon 28.6 $0.41
Muse Wonder 27.6 $1.13
Muse Ember 25.7 $0.06
Gemini 3.8 Flash 24.3 $0.27
Grok 4.7 23.0 $0.63
Gemini 4 Carbon 22.0 $0.30
Muse Spark 1.3 21.7 $0.42
GLM-5.3 21.2 $0.10
DeepSeek V4.1 Flash 19.8 $0.05
GPT-6 Luna 18.1 $0.02
Gemma 4 31B IT 17.1 $0.08
Kimi K3.1 Pro 15.5 $0.47
Muse Spark Open 15.2 $0.42
Claude Haiku 5.5 15.0 $0.03
Nemotron 3.5 Super VL 11.2 $0.11
Qwen4-Max 11.0 $0.16
Qwen 3.8 Max 8.68 $0.13

2 models have held number one on the official leaderboard since its release in November 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard 2 leaders

Source: Official KradleBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard
0.00 20.0 40.0 60.0 80.0 2026 2027 0.00 20.0 40.0 60.0 80.0 2026 2027

Source: Official KradleBench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Games and long tasks

All 261 in Games and long tasks
  • ARC-AGI-3 In use Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. GPT-6.2 Astra 100%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Each game scores in-game progress or outcomes, and the board ranks models by their average rank across the games. Gemini 2.5 Pro 3.90
  • MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks. Seed2.0 Pro 49%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. The score is harmful actions per 100 steps, where lower is better. The board also reports game progress. Llama 4 Scout 7.00
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. GPT-6.1 Terra 76.8%
  • Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (26 entries).
  • Pass Models scored 26 models we track.
  • Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
  • Pass Frontier models 9 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on kradle.ai: 26 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check