KUMO
Agents make queries and use the responses to identify hidden facts in generated reasoning games. Harbor offers 5,050 easy games and 250 hard games.
The 212-task Harbor comparison available November 19, 2026 tests GPT-5 mini and GPT-5 nano. GPT-5 mini with Codex led at 92.9% accuracy.
- Made by
- KUMO team
- Released
- April 2025
- Kind
- Puzzles and chess
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- KUMO
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
-
DeepSeek-R1 #130 on the capability index
- Last model added
- March 2025
- Scores refreshed
- November 20th, 2026
QwQ-32B leads with 86.3% among models released by March 2025 QwQ-32B leads with 86.3% among models released by March 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in March 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 86.3
QwQ-32B
- 86.0 deepseek-v3 (snapshot unspecified)
- 83.0
DeepSeek-R1
- 75.7
Mistral Large 2
- 73.3 qwen2.5-32b-instruct
- 72.7
Qwen2.5-72B
- 71.5 qwen2.5-14b-instruct
- 63.7
Llama 3.3 70B Instruct
- 62.7 llama-3.1-70b-instruct
- 44.5 gemma-2-27b-it
- 34.7 qwen2.5-7b-instruct
- 33.4 llama-3.1-8b-instruct
- 26.4 gemma-2-9b-it
- 16.6 qwen2.5-3b-instruct
- 15.6 llama-3.2-3b-instruct
- 1
QwQ-32B 86.3%
- 2 deepseek-v3 (snapshot unspecified) 86%
- 3
DeepSeek-R1 83%
- 4
Mistral Large 2 Nov 2024 75.7%
- 5 qwen2.5-32b-instruct 73.3%
- 6
Qwen2.5-72B 72.7%
- 7 qwen2.5-14b-instruct 71.5%
- 8
Llama 3.3 70B Instruct 63.7%
- 9 llama-3.1-70b-instruct 62.7%
- 10 gemma-2-27b-it 44.5%
- 11 qwen2.5-7b-instruct 34.7%
- 12 llama-3.1-8b-instruct 33.4%
- 13 gemma-2-9b-it 26.4%
- 14 qwen2.5-3b-instruct 16.6%
- 15 llama-3.2-3b-instruct 15.6%
Source: Official KUMO leaderboard
- 1
QwQ-32B 86.3%
- 2
DeepSeek-R1 83%
- 3
Mistral Large 2 Nov 2024 75.7%
- 4
Qwen2.5-72B 72.7%
- 5
Llama 3.3 70B Instruct 63.7%
Source: Official KUMO leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 86.3% | Mar 2025 | Official leaderboard | |
| 2 | deepseek-v3 (snapshot unspecified) | 86% | Official leaderboard | ||
| 3 | | 83% | Jan 2025 | Official leaderboard | |
| 4 | | 75.7% | Nov 2024 | Official leaderboard | |
| 5 | qwen2.5-32b-instruct | 73.3% | Official leaderboard | ||
| 6 | | 72.7% | Sep 2024 | Official leaderboard | |
| 7 | qwen2.5-14b-instruct | 71.5% | Official leaderboard | ||
| 8 | | 63.7% | Dec 2024 | Official leaderboard | |
| 9 | llama-3.1-70b-instruct | 62.7% | Official leaderboard | ||
| 10 | gemma-2-27b-it | 44.5% | Official leaderboard | ||
| 11 | qwen2.5-7b-instruct | 34.7% | Official leaderboard | ||
| 12 | llama-3.1-8b-instruct | 33.4% | Official leaderboard | ||
| 13 | gemma-2-9b-it | 26.4% | Official leaderboard | ||
| 14 | qwen2.5-3b-instruct | 16.6% | Official leaderboard | ||
| 15 | llama-3.2-3b-instruct | 15.6% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 86.3% | Mar 2025 | Official leaderboard | |
| 2 | | 83% | Jan 2025 | Official leaderboard | |
| 3 | | 75.7% | Nov 2024 | Official leaderboard | |
| 4 | | 72.7% | Sep 2024 | Official leaderboard | |
| 5 | | 63.7% | Dec 2024 | Official leaderboard |
Rankings by part
We read 100 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as RelicEnv, HerbEnv and TransdimensionalEnv. QwQ-32B leads 69 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
RelicEnv
5 models we track have a score here.
- 1
QwQ-32B 95.6%
- 2
DeepSeek-R1 91.6%
- 3
Qwen2.5-72B ↑1 89.2%
- 4
Mistral Large 2 Nov 2024 ↓1 85.6%
- 5
Llama 3.3 70B Instruct 83.6%
HerbEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 93.6%
- 2
QwQ-32B ↓1 92.4%
- 3
Qwen2.5-72B ↑1 80.4%
- 4
Mistral Large 2 Nov 2024 ↓1 76%
- 5
Llama 3.3 70B Instruct 61.2%
TransdimensionalEnv
5 models we track have a score here.
- 1
QwQ-32B 96.8%
- 2
DeepSeek-R1 93.6%
- 3
Mistral Large 2 Nov 2024 86%
- 4
Llama 3.3 70B Instruct ↑1 86%
- 5
Qwen2.5-72B ↓1 83.6%
SorcererEnv
5 models we track have a score here.
- 1
QwQ-32B 84%
- 2
Qwen2.5-72B ↑2 83.2%
- 3
DeepSeek-R1 ↓1 82.4%
- 4
Mistral Large 2 Nov 2024 ↓1 80%
- 5
Llama 3.3 70B Instruct 73.6%
QuantumEnv
5 models we track have a score here.
- 1
Qwen2.5-72B ↑3 91.6%
- 2
Mistral Large 2 Nov 2024 ↑1 87.2%
- 3
QwQ-32B ↓2 86.8%
- 4
DeepSeek-R1 ↓2 86.4%
- 5
Llama 3.3 70B Instruct 84.8%
AstronomyEnv
5 models we track have a score here.
- 1
QwQ-32B 85.2%
- 2
DeepSeek-R1 82%
- 3
Mistral Large 2 Nov 2024 80%
- 4
Qwen2.5-72B 74.8%
- 5
Llama 3.3 70B Instruct 65.2%
MusicGenresEnv
5 models we track have a score here.
- 1
QwQ-32B 87.6%
- 2
DeepSeek-R1 79.2%
- 3
Mistral Large 2 Nov 2024 73.2%
- 4
Qwen2.5-72B 69.6%
- 5
Llama 3.3 70B Instruct 59.2%
CloudEnv
5 models we track have a score here.
- 1
QwQ-32B 91.2%
- 2
DeepSeek-R1 90.8%
- 3
Mistral Large 2 Nov 2024 83.6%
- 4
Qwen2.5-72B 71.2%
- 5
Llama 3.3 70B Instruct 69.6%
CuisineEnv
5 models we track have a score here.
- 1
QwQ-32B 88%
- 2
DeepSeek-R1 79.6%
- 3
Qwen2.5-72B ↑1 65.6%
- 4
Mistral Large 2 Nov 2024 ↓1 64.4%
- 5
Llama 3.3 70B Instruct 44.4%
PlantEnv
5 models we track have a score here.
- 1
QwQ-32B 54.8%
- 2
DeepSeek-R1 50%
- 3
Mistral Large 2 Nov 2024 26.4%
- 4
Qwen2.5-72B 22.8%
- 5
Llama 3.3 70B Instruct 7.60%
HistoricalEnv
5 models we track have a score here.
- 1
QwQ-32B 88.4%
- 2
DeepSeek-R1 82.8%
- 3
Mistral Large 2 Nov 2024 68.8%
- 4
Llama 3.3 70B Instruct ↑1 65.2%
- 5
Qwen2.5-72B ↓1 51.2%
GadgetEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 93.2%
- 2
QwQ-32B ↓1 93.2%
- 3
Qwen2.5-72B ↑1 89.2%
- 4
Mistral Large 2 Nov 2024 ↓1 88.4%
- 5
Llama 3.3 70B Instruct 73.6%
TimeTravelEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 93.2%
- 2
QwQ-32B ↓1 92.4%
- 3
Mistral Large 2 Nov 2024 88.4%
- 4
Qwen2.5-72B 80.4%
- 5
Llama 3.3 70B Instruct 71.6%
PollutionEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 85.6%
- 2
QwQ-32B ↓1 85.2%
- 3
Mistral Large 2 Nov 2024 78.4%
- 4
Qwen2.5-72B 70.4%
- 5
Llama 3.3 70B Instruct 69.6%
DemographicEnv
5 models we track have a score here.
- 1
QwQ-32B 89.6%
- 2
DeepSeek-R1 87.6%
- 3
Mistral Large 2 Nov 2024 82%
- 4
Qwen2.5-72B 77.2%
- 5
Llama 3.3 70B Instruct 74.8%
GeneticEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 95.2%
- 2
Qwen2.5-72B ↑2 94%
- 3
QwQ-32B ↓2 93.2%
- 4
Llama 3.3 70B Instruct ↑1 91.6%
- 5
Mistral Large 2 Nov 2024 ↓2 90.4%
CraftsmanEnv
5 models we track have a score here.
- 1
QwQ-32B 81.6%
- 2
DeepSeek-R1 78.8%
- 3
Mistral Large 2 Nov 2024 70.8%
- 4
Qwen2.5-72B 66.4%
- 5
Llama 3.3 70B Instruct 41.6%
StarConstellationEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 74.8%
- 2
QwQ-32B ↓1 73.6%
- 3
Mistral Large 2 Nov 2024 61.2%
- 4
Qwen2.5-72B 58.4%
- 5
Llama 3.3 70B Instruct 41.2%
MythicalCreatureEnv
5 models we track have a score here.
- 1
QwQ-32B 85.2%
- 2
DeepSeek-R1 84%
- 3
Mistral Large 2 Nov 2024 74.8%
- 4
Qwen2.5-72B 66.8%
- 5
Llama 3.3 70B Instruct 64.8%
ArtStyleEnv
5 models we track have a score here.
- 1
QwQ-32B 86.8%
- 2
Mistral Large 2 Nov 2024 ↑1 82.8%
- 3
DeepSeek-R1 ↓1 82%
- 4
Qwen2.5-72B 74.8%
- 5
Llama 3.3 70B Instruct 62%
CookingEnv
5 models we track have a score here.
- 1
QwQ-32B 90%
- 2
DeepSeek-R1 81.2%
- 3
Mistral Large 2 Nov 2024 74%
- 4
Qwen2.5-72B 70%
- 5
Llama 3.3 70B Instruct 65.6%
HistoricalBattleEnv
5 models we track have a score here.
- 1
QwQ-32B 85.6%
- 2
DeepSeek-R1 78.8%
- 3
Llama 3.3 70B Instruct ↑2 62%
- 4
Mistral Large 2 Nov 2024 ↓1 60.8%
- 5
Qwen2.5-72B ↓1 42.4%
FungalEnv
5 models we track have a score here.
- 1
QwQ-32B 78.8%
- 2
DeepSeek-R1 76.4%
- 3
Mistral Large 2 Nov 2024 74.4%
- 4
Qwen2.5-72B 68.4%
- 5
Llama 3.3 70B Instruct 64.4%
CryptographyEnv
5 models we track have a score here.
- 1
QwQ-32B 84.8%
- 2
DeepSeek-R1 74%
- 3
Mistral Large 2 Nov 2024 68%
- 4
Qwen2.5-72B 52.8%
- 5
Llama 3.3 70B Instruct 51.2%
StorageEnv
5 models we track have a score here.
- 1
QwQ-32B 94.4%
- 2
DeepSeek-R1 91.2%
- 3
Mistral Large 2 Nov 2024 88.8%
- 4
Qwen2.5-72B 81.2%
- 5
Llama 3.3 70B Instruct 79.6%
RoverEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 90%
- 2
QwQ-32B ↓1 87.2%
- 3
Mistral Large 2 Nov 2024 82.8%
- 4
Qwen2.5-72B 79.6%
- 5
Llama 3.3 70B Instruct 72.4%
FashionEnv
5 models we track have a score here.
- 1
QwQ-32B 86%
- 2
Qwen2.5-72B ↑2 76.8%
- 3
Mistral Large 2 Nov 2024 75.6%
- 4
DeepSeek-R1 ↓2 75.6%
- 5
Llama 3.3 70B Instruct 61.6%
LicenseEnv
5 models we track have a score here.
- 1
QwQ-32B 58%
- 2
DeepSeek-R1 52%
- 3
Mistral Large 2 Nov 2024 50.4%
- 4
Qwen2.5-72B 48.4%
- 5
Llama 3.3 70B Instruct 46%
VirusClassificationEnv
5 models we track have a score here.
- 1
QwQ-32B 68.4%
- 2
DeepSeek-R1 64%
- 3
Mistral Large 2 Nov 2024 53.2%
- 4
Qwen2.5-72B 42.8%
- 5
Llama 3.3 70B Instruct 39.6%
TestingEnv
5 models we track have a score here.
- 1
QwQ-32B 80%
- 2
Mistral Large 2 Nov 2024 ↑1 76.4%
- 3
DeepSeek-R1 ↓1 76.4%
- 4
Qwen2.5-72B 70.8%
- 5
Llama 3.3 70B Instruct 70.4%
NarrativeDetectEnv
5 models we track have a score here.
- 1
Mistral Large 2 Nov 2024 ↑2 79.2%
- 2
QwQ-32B ↓1 78.4%
- 3
Qwen2.5-72B ↑1 76%
- 4
DeepSeek-R1 ↓2 74.8%
- 5
Llama 3.3 70B Instruct 70.4%
RenewableEnergyEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 98%
- 2
QwQ-32B ↓1 98%
- 3
Qwen2.5-72B ↑1 88.8%
- 4
Mistral Large 2 Nov 2024 ↓1 85.6%
- 5
Llama 3.3 70B Instruct 82.4%
CelestialEnv
5 models we track have a score here.
- 1
QwQ-32B 88.8%
- 2
DeepSeek-R1 83.6%
- 3
Mistral Large 2 Nov 2024 82%
- 4
Llama 3.3 70B Instruct ↑1 74.4%
- 5
Qwen2.5-72B ↓1 72.4%
SpiceEnv
5 models we track have a score here.
- 1
QwQ-32B 87.2%
- 2
DeepSeek-R1 76.8%
- 3
Qwen2.5-72B ↑1 58.8%
- 4
Mistral Large 2 Nov 2024 ↓1 58%
- 5
Llama 3.3 70B Instruct 33.6%
WildlifeEnv
5 models we track have a score here.
- 1
QwQ-32B 71.6%
- 2
Mistral Large 2 Nov 2024 ↑1 62.8%
- 3
Qwen2.5-72B ↑1 61.6%
- 4
Llama 3.3 70B Instruct ↑1 61.6%
- 5
DeepSeek-R1 ↓3 60.4%
VehicleEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 92%
- 2
QwQ-32B ↓1 91.6%
- 3
Mistral Large 2 Nov 2024 88%
- 4
Qwen2.5-72B 78%
- 5
Llama 3.3 70B Instruct 69.6%
BeverageEnv
5 models we track have a score here.
- 1
QwQ-32B 84.4%
- 2
DeepSeek-R1 75.2%
- 3
Mistral Large 2 Nov 2024 74.8%
- 4
Qwen2.5-72B 72.4%
- 5
Llama 3.3 70B Instruct 58%
ControlEnv
5 models we track have a score here.
- 1
QwQ-32B 93.2%
- 2
DeepSeek-R1 91.2%
- 3
Qwen2.5-72B ↑1 84%
- 4
Mistral Large 2 Nov 2024 ↓1 78.4%
- 5
Llama 3.3 70B Instruct 63.2%
CurrencyEnv
5 models we track have a score here.
- 1
QwQ-32B 96.4%
- 2
DeepSeek-R1 93.2%
- 3
Mistral Large 2 Nov 2024 89.6%
- 4
Qwen2.5-72B 83.6%
- 5
Llama 3.3 70B Instruct 70%
MarketingEnv
5 models we track have a score here.
- 1
QwQ-32B 83.2%
- 2
DeepSeek-R1 79.6%
- 3
Llama 3.3 70B Instruct ↑2 76.4%
- 4
Mistral Large 2 Nov 2024 ↓1 76%
- 5
Qwen2.5-72B ↓1 73.2%
BotanicalEnv
5 models we track have a score here.
- 1
QwQ-32B 96.8%
- 2
DeepSeek-R1 94%
- 3
Qwen2.5-72B ↑1 90.4%
- 4
Mistral Large 2 Nov 2024 ↓1 86.4%
- 5
Llama 3.3 70B Instruct 84.8%
CircusActEnv
5 models we track have a score here.
- 1
QwQ-32B 87.2%
- 2
DeepSeek-R1 82%
- 3
Qwen2.5-72B ↑1 76.8%
- 4
Mistral Large 2 Nov 2024 ↓1 74.8%
- 5
Llama 3.3 70B Instruct 62%
AudioDialectEnv
5 models we track have a score here.
- 1
QwQ-32B 81.2%
- 2
DeepSeek-R1 79.2%
- 3
Mistral Large 2 Nov 2024 68%
- 4
Qwen2.5-72B 52.8%
- 5
Llama 3.3 70B Instruct 32.4%
LeadershipEnv
5 models we track have a score here.
- 1
QwQ-32B 82.8%
- 2
Qwen2.5-72B ↑2 76.4%
- 3
DeepSeek-R1 ↓1 74.8%
- 4
Mistral Large 2 Nov 2024 ↓1 72.8%
- 5
Llama 3.3 70B Instruct 69.2%
TransportEnv
5 models we track have a score here.
- 1
QwQ-32B 92%
- 2
DeepSeek-R1 82.4%
- 3
Qwen2.5-72B ↑1 80%
- 4
Mistral Large 2 Nov 2024 ↓1 78.8%
- 5
Llama 3.3 70B Instruct 60%
EcologicalEnv
5 models we track have a score here.
- 1
QwQ-32B 88.8%
- 2
DeepSeek-R1 87.2%
- 3
Qwen2.5-72B ↑1 79.2%
- 4
Mistral Large 2 Nov 2024 ↓1 77.2%
- 5
Llama 3.3 70B Instruct 64%
MythicEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 87.2%
- 2
QwQ-32B ↓1 84%
- 3
Mistral Large 2 Nov 2024 69.6%
- 4
Qwen2.5-72B 67.2%
- 5
Llama 3.3 70B Instruct 57.6%
EnzymeEnv
5 models we track have a score here.
- 1
QwQ-32B 80.4%
- 2
DeepSeek-R1 76%
- 3
Mistral Large 2 Nov 2024 73.2%
- 4
Qwen2.5-72B 67.6%
- 5
Llama 3.3 70B Instruct 56.4%
OSKernelEnv
5 models we track have a score here.
- 1
QwQ-32B 98.4%
- 2
DeepSeek-R1 96%
- 3
Mistral Large 2 Nov 2024 85.6%
- 4
Qwen2.5-72B 78.8%
- 5
Llama 3.3 70B Instruct 74%
MineralClassificationEnv
5 models we track have a score here.
- 1
QwQ-32B 96.4%
- 2
DeepSeek-R1 92%
- 3
Mistral Large 2 Nov 2024 90.8%
- 4
Llama 3.3 70B Instruct ↑1 89.6%
- 5
Qwen2.5-72B ↓1 88.4%
EconomicEnv
5 models we track have a score here.
- 1
QwQ-32B 97.2%
- 2
DeepSeek-R1 93.6%
- 3
Mistral Large 2 Nov 2024 92.4%
- 4
Qwen2.5-72B 89.6%
- 5
Llama 3.3 70B Instruct 82.4%
DetectiveEnv
5 models we track have a score here.
- 1
QwQ-32B 98.4%
- 2
DeepSeek-R1 96.4%
- 3
Mistral Large 2 Nov 2024 89.6%
- 4
Qwen2.5-72B 84.8%
- 5
Llama 3.3 70B Instruct 79.2%
ChessEnv
5 models we track have a score here.
- 1
Mistral Large 2 Nov 2024 ↑2 74.4%
- 2
QwQ-32B ↓1 66.4%
- 3
DeepSeek-R1 ↓1 65.2%
- 4
Llama 3.3 70B Instruct ↑1 62.8%
- 5
Qwen2.5-72B ↓1 57.2%
MythicalEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 89.6%
- 2
QwQ-32B ↓1 87.6%
- 3
Mistral Large 2 Nov 2024 72.8%
- 4
Qwen2.5-72B 63.2%
- 5
Llama 3.3 70B Instruct 55.6%
ChemicalCompoundsEnv
5 models we track have a score here.
- 1
QwQ-32B 75.2%
- 2
DeepSeek-R1 62.4%
- 3
Mistral Large 2 Nov 2024 43.2%
- 4
Qwen2.5-72B 28.4%
- 5
Llama 3.3 70B Instruct 18%
ArchitecturalEnv
5 models we track have a score here.
- 1
QwQ-32B 82.4%
- 2
DeepSeek-R1 74.4%
- 3
Mistral Large 2 Nov 2024 72.4%
- 4
Qwen2.5-72B 71.2%
- 5
Llama 3.3 70B Instruct 50.8%
ComputationEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 94%
- 2
QwQ-32B ↓1 90.8%
- 3
Qwen2.5-72B ↑1 85.6%
- 4
Llama 3.3 70B Instruct ↑1 85.6%
- 5
Mistral Large 2 Nov 2024 ↓2 82.8%
MachinePartEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 95.2%
- 2
QwQ-32B ↓1 91.6%
- 3
Llama 3.3 70B Instruct ↑2 89.6%
- 4
Mistral Large 2 Nov 2024 ↓1 87.6%
- 5
Qwen2.5-72B ↓1 82.8%
LiteraryEnv
5 models we track have a score here.
- 1
QwQ-32B 96.8%
- 2
DeepSeek-R1 91.2%
- 3
Qwen2.5-72B ↑1 90%
- 4
Mistral Large 2 Nov 2024 ↓1 88.4%
- 5
Llama 3.3 70B Instruct 66.4%
MarineEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 90%
- 2
QwQ-32B ↓1 88.8%
- 3
Qwen2.5-72B ↑1 83.2%
- 4
Mistral Large 2 Nov 2024 ↓1 78.8%
- 5
Llama 3.3 70B Instruct 70.4%
PhilosophyEnv
5 models we track have a score here.
- 1
Qwen2.5-72B ↑3 87.2%
- 2
QwQ-32B ↓1 83.2%
- 3
Llama 3.3 70B Instruct ↑2 79.6%
- 4
Mistral Large 2 Nov 2024 ↓1 78.8%
- 5
DeepSeek-R1 ↓3 78%
ArchaeologicalEnv
5 models we track have a score here.
- 1
QwQ-32B 75.6%
- 2
DeepSeek-R1 70.4%
- 3
Mistral Large 2 Nov 2024 64%
- 4
Qwen2.5-72B 56.4%
- 5
Llama 3.3 70B Instruct 54.8%
GemstoneEnv
5 models we track have a score here.
- 1
QwQ-32B 63.6%
- 2
DeepSeek-R1 56.8%
- 3
Mistral Large 2 Nov 2024 52.8%
- 4
Llama 3.3 70B Instruct ↑1 45.6%
- 5
Qwen2.5-72B ↓1 44%
MicrobiologyEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 95.2%
- 2
QwQ-32B ↓1 93.2%
- 3
Qwen2.5-72B ↑1 84.4%
- 4
Mistral Large 2 Nov 2024 ↓1 76.4%
- 5
Llama 3.3 70B Instruct 49.6%
SciFiEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 92%
- 2
QwQ-32B ↓1 90%
- 3
Qwen2.5-72B ↑1 78.8%
- 4
Mistral Large 2 Nov 2024 ↓1 73.6%
- 5
Llama 3.3 70B Instruct 58.8%
HormoneEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 94.4%
- 2
QwQ-32B ↓1 85.2%
- 3
Llama 3.3 70B Instruct ↑2 84.8%
- 4
Mistral Large 2 Nov 2024 ↓1 80%
- 5
Qwen2.5-72B ↓1 76.4%
SculptorEnv
5 models we track have a score here.
- 1
QwQ-32B 81.2%
- 2
DeepSeek-R1 74.8%
- 3
Mistral Large 2 Nov 2024 68.4%
- 4
Qwen2.5-72B 60%
- 5
Llama 3.3 70B Instruct 53.2%
NeuroEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 93.2%
- 2
Llama 3.3 70B Instruct ↑3 92.8%
- 3
Qwen2.5-72B ↑1 88.8%
- 4
Mistral Large 2 Nov 2024 ↓1 88.8%
- 5
QwQ-32B ↓4 85.2%
OceanEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 83.6%
- 2
QwQ-32B ↓1 82.4%
- 3
Mistral Large 2 Nov 2024 65.6%
- 4
Qwen2.5-72B 62%
- 5
Llama 3.3 70B Instruct 47.6%
MineralEnv
5 models we track have a score here.
- 1
QwQ-32B 87.2%
- 2
DeepSeek-R1 82%
- 3
Qwen2.5-72B ↑1 68.4%
- 4
Mistral Large 2 Nov 2024 ↓1 66%
- 5
Llama 3.3 70B Instruct 56%
FishEnv
5 models we track have a score here.
- 1
QwQ-32B 92.4%
- 2
DeepSeek-R1 86.8%
- 3
Mistral Large 2 Nov 2024 73.6%
- 4
Qwen2.5-72B 72%
- 5
Llama 3.3 70B Instruct 61.6%
MartialArtsEnv
5 models we track have a score here.
- 1
QwQ-32B 75.2%
- 2
DeepSeek-R1 71.6%
- 3
Mistral Large 2 Nov 2024 56.8%
- 4
Qwen2.5-72B 56%
- 5
Llama 3.3 70B Instruct 52%
RocketFuelEnv
5 models we track have a score here.
- 1
QwQ-32B 90.4%
- 2
DeepSeek-R1 89.6%
- 3
Mistral Large 2 Nov 2024 82.8%
- 4
Qwen2.5-72B 81.6%
- 5
Llama 3.3 70B Instruct 72.4%
MLEnv
5 models we track have a score here.
- 1
Qwen2.5-72B ↑3 79.2%
- 2
QwQ-32B ↓1 77.2%
- 3
Mistral Large 2 Nov 2024 76.4%
- 4
DeepSeek-R1 ↓2 65.2%
- 5
Llama 3.3 70B Instruct 59.6%
PoliticalManifestoEnv
5 models we track have a score here.
- 1
QwQ-32B 88%
- 2
Mistral Large 2 Nov 2024 ↑1 79.6%
- 3
DeepSeek-R1 ↓1 79.2%
- 4
Qwen2.5-72B 78.4%
- 5
Llama 3.3 70B Instruct 69.2%
CoffeeEnv
5 models we track have a score here.
- 1
QwQ-32B 93.6%
- 2
DeepSeek-R1 92.8%
- 3
Mistral Large 2 Nov 2024 82.8%
- 4
Qwen2.5-72B 76.4%
- 5
Llama 3.3 70B Instruct 66%
MotifAnalysisEnv
5 models we track have a score here.
- 1
QwQ-32B 86.4%
- 2
DeepSeek-R1 82.4%
- 3
Qwen2.5-72B ↑1 59.2%
- 4
Mistral Large 2 Nov 2024 ↓1 46.4%
- 5
Llama 3.3 70B Instruct 22.4%
NutritionEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 94.4%
- 2
QwQ-32B ↓1 91.2%
- 3
Qwen2.5-72B ↑1 87.6%
- 4
Mistral Large 2 Nov 2024 ↓1 83.2%
- 5
Llama 3.3 70B Instruct 70.4%
MalwareEnv
5 models we track have a score here.
- 1
QwQ-32B 91.6%
- 2
DeepSeek-R1 90%
- 3
Mistral Large 2 Nov 2024 75.2%
- 4
Qwen2.5-72B 72%
- 5
Llama 3.3 70B Instruct 54.8%
GeologicalEnv
5 models we track have a score here.
- 1
QwQ-32B 88.8%
- 2
DeepSeek-R1 84.8%
- 3
Mistral Large 2 Nov 2024 81.2%
- 4
Qwen2.5-72B 67.6%
- 5
Llama 3.3 70B Instruct 50.8%
TheatricalEnv
5 models we track have a score here.
- 1
QwQ-32B 90.8%
- 2
Mistral Large 2 Nov 2024 ↑1 84.4%
- 3
DeepSeek-R1 ↓1 82.4%
- 4
Qwen2.5-72B 80.8%
- 5
Llama 3.3 70B Instruct 76.8%
PrintingTechniqueEnv
5 models we track have a score here.
- 1
QwQ-32B 92.8%
- 2
DeepSeek-R1 85.2%
- 3
Qwen2.5-72B ↑1 77.6%
- 4
Mistral Large 2 Nov 2024 ↓1 70.4%
- 5
Llama 3.3 70B Instruct 54.8%
StellarEnv
5 models we track have a score here.
- 1
QwQ-32B 77.6%
- 2
Mistral Large 2 Nov 2024 ↑1 73.2%
- 3
Qwen2.5-72B ↑1 69.6%
- 4
DeepSeek-R1 ↓2 64.8%
- 5
Llama 3.3 70B Instruct 62.4%
SoilEnv
5 models we track have a score here.
- 1
QwQ-32B 84.8%
- 2
Qwen2.5-72B ↑2 82.4%
- 3
DeepSeek-R1 ↓1 80.4%
- 4
Mistral Large 2 Nov 2024 ↓1 76%
- 5
Llama 3.3 70B Instruct 66.4%
SoftwareEnv
5 models we track have a score here.
- 1
QwQ-32B 88%
- 2
Qwen2.5-72B ↑2 84%
- 3
DeepSeek-R1 ↓1 83.6%
- 4
Mistral Large 2 Nov 2024 ↓1 80.4%
- 5
Llama 3.3 70B Instruct 78.4%
CarIdentificationEnv
5 models we track have a score here.
- 1
QwQ-32B 98.8%
- 2
DeepSeek-R1 95.2%
- 3
Qwen2.5-72B ↑1 93.6%
- 4
Mistral Large 2 Nov 2024 ↓1 91.2%
- 5
Llama 3.3 70B Instruct 85.2%
PharmaceuticalEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 86.4%
- 2
QwQ-32B ↓1 80%
- 3
Qwen2.5-72B ↑1 75.6%
- 4
Mistral Large 2 Nov 2024 ↓1 75.6%
- 5
Llama 3.3 70B Instruct 47.2%
NetworkEnv
5 models we track have a score here.
- 1
QwQ-32B 82.8%
- 2
Mistral Large 2 Nov 2024 ↑1 78%
- 3
DeepSeek-R1 ↓1 73.6%
- 4
Qwen2.5-72B 71.6%
- 5
Llama 3.3 70B Instruct 70.4%
BirdNestEnv
5 models we track have a score here.
- 1
QwQ-32B 81.2%
- 2
DeepSeek-R1 75.6%
- 3
Mistral Large 2 Nov 2024 49.2%
- 4
Qwen2.5-72B 42.4%
- 5
Llama 3.3 70B Instruct 20.8%
EnergyEnv
5 models we track have a score here.
- 1
QwQ-32B 88.8%
- 2
DeepSeek-R1 87.2%
- 3
Mistral Large 2 Nov 2024 78%
- 4
Qwen2.5-72B 58.8%
- 5
Llama 3.3 70B Instruct 39.6%
LanguageEnv
5 models we track have a score here.
- 1
QwQ-32B 81.2%
- 2
DeepSeek-R1 71.6%
- 3
Qwen2.5-72B ↑1 56%
- 4
Mistral Large 2 Nov 2024 ↓1 53.6%
- 5
Llama 3.3 70B Instruct 33.6%
AlgorithmEnv
5 models we track have a score here.
- 1
QwQ-32B 81.2%
- 2
Mistral Large 2 Nov 2024 ↑1 78.8%
- 3
DeepSeek-R1 ↓1 72.4%
- 4
Qwen2.5-72B 66%
- 5
Llama 3.3 70B Instruct 48.4%
MathematicalEnv
5 models we track have a score here.
- 1
Mistral Large 2 Nov 2024 ↑2 88.4%
- 2
DeepSeek-R1 87.6%
- 3
Llama 3.3 70B Instruct ↑2 84%
- 4
QwQ-32B ↓3 81.6%
- 5
Qwen2.5-72B ↓1 79.2%
MusicalEnv
5 models we track have a score here.
- 1
QwQ-32B 90.4%
- 2
DeepSeek-R1 89.2%
- 3
Mistral Large 2 Nov 2024 88.4%
- 4
Qwen2.5-72B 84%
- 5
Llama 3.3 70B Instruct 82.4%
InventorEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 90.8%
- 2
QwQ-32B ↓1 90%
- 3
Mistral Large 2 Nov 2024 80.4%
- 4
Qwen2.5-72B 77.2%
- 5
Llama 3.3 70B Instruct 61.6%
MedicalEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 93.6%
- 2
QwQ-32B ↓1 92%
- 3
Qwen2.5-72B ↑1 89.6%
- 4
Mistral Large 2 Nov 2024 ↓1 89.6%
- 5
Llama 3.3 70B Instruct 89.6%
MusicEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 86.8%
- 2
QwQ-32B ↓1 86.8%
- 3
Qwen2.5-72B ↑1 75.6%
- 4
Mistral Large 2 Nov 2024 ↓1 66.4%
- 5
Llama 3.3 70B Instruct 59.6%
FantasyEnv
5 models we track have a score here.
- 1
DeepSeek-R1 ↑1 97.2%
- 2
QwQ-32B ↓1 97.2%
- 3
Mistral Large 2 Nov 2024 82.4%
- 4
Llama 3.3 70B Instruct ↑1 70.4%
- 5
Qwen2.5-72B ↓1 57.2%
EducationEnv
5 models we track have a score here.
- 1
QwQ-32B 93.6%
- 2
Llama 3.3 70B Instruct ↑3 90.4%
- 3
DeepSeek-R1 ↓1 90%
- 4
Mistral Large 2 Nov 2024 ↓1 87.6%
- 5
Qwen2.5-72B ↓1 84.8%
ChemicalEnv
5 models we track have a score here.
- 1
QwQ-32B 82%
- 2
Qwen2.5-72B ↑2 72%
- 3
Mistral Large 2 Nov 2024 68%
- 4
DeepSeek-R1 ↓2 68%
- 5
Llama 3.3 70B Instruct 46%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official KUMO leaderboard
QwQ-32B on the official leaderboard passed 85% in March 2025
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official KUMO 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
- 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
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the official leaderboard (15 entries).
- Pass Models scored 5 models we track, plus 10 other entries on its leaderboard.
- Note Newest score The newest came 20 months ago, in March 2025, for QwQ-32B. That’s expected for a saturated test.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is DeepSeek-R1 (#130).
- Note Headroom Saturated. The best is 86.3%, past 85% since March 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 5 rows for models we track and 10 other entries 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.