Benchmark Almanac

KUMO

QwQ-32B 86.3% leads since April 2025
Not in model cards

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
In and out
Text → Actions, Text
Interaction
Agentic
Environment
KUMO
Items from
Generated
Refresh
One fixed set
Run by
Anyone
In views
Games, Agents
Strongest model tested
DeepSeek-R1 #130 on the capability index
Last model added
March 2025
Scores refreshed
November 20th, 2026
15
Entries on its own leaderboard, 5 models we track

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:

  1. 1 QwQ-32B 86.3%
  2. 2 deepseek-v3 (snapshot unspecified) 86%
  3. 3 DeepSeek-R1 83%
  4. 4 Mistral Large 2 Nov 2024 75.7%
  5. 5 qwen2.5-32b-instruct 73.3%
  6. 6 Qwen2.5-72B 72.7%
  7. 7 qwen2.5-14b-instruct 71.5%
  8. 8 Llama 3.3 70B Instruct 63.7%
  9. 9 llama-3.1-70b-instruct 62.7%
  10. 10 gemma-2-27b-it 44.5%
  11. 11 qwen2.5-7b-instruct 34.7%
  12. 12 llama-3.1-8b-instruct 33.4%
  13. 13 gemma-2-9b-it 26.4%
  14. 14 qwen2.5-3b-instruct 16.6%
  15. 15 llama-3.2-3b-instruct 15.6%

Source: Official KUMO leaderboard

#EntryScoreSettingDateSource
1 QwQ-32B 86.3% Mar 2025 Official leaderboard
2 deepseek-v3 (snapshot unspecified) 86% Official leaderboard
3 DeepSeek-R1 83% Jan 2025 Official leaderboard
4 Mistral Large 2 Nov 2024 75.7% Nov 2024 Official leaderboard
5 qwen2.5-32b-instruct 73.3% Official leaderboard
6 Qwen2.5-72B 72.7% Sep 2024 Official leaderboard
7 qwen2.5-14b-instruct 71.5% Official leaderboard
8 Llama 3.3 70B Instruct 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
#ModelScoreSettingDateSource
1 QwQ-32B 86.3% Mar 2025 Official leaderboard
2 DeepSeek-R1 83% Jan 2025 Official leaderboard
3 Mistral Large 2 Nov 2024 75.7% Nov 2024 Official leaderboard
4 Qwen2.5-72B 72.7% Sep 2024 Official leaderboard
5 Llama 3.3 70B Instruct 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. 1 QwQ-32B 95.6%
  2. 2 DeepSeek-R1 91.6%
  3. 3 Qwen2.5-72B ↑1 89.2%
  4. 4 Mistral Large 2 Nov 2024 ↓1 85.6%
  5. 5 Llama 3.3 70B Instruct 83.6%

Number one over time

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

Qwen
Own leaderboard Passed 85% in Apr 2025, by QwQ-32B
QwQ-32B 86.3%

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.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) QwQ-32B Mistral Large 2 Qwen2.5-72B 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) QwQ-32B Mistral Large 2

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

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

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.

How we check