Benchmark Almanac

ProxyWar

DeepSeek-R1 May 2025 39.6% leads since February 2026
Not in model cards

Models write bots for nine games. The harness tests each bot and runs tournaments to measure wins, errors, and decision time.

DeepSeek R1 0528 led 18 models with a 39.6% overall win rate in February 2026. The field included GPT, Claude, Gemini, and Qwen models.

Released
February 2026
Kind
Multi-agent games
Level
University
Scored
Ranked by matches
Answers
Code
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Code → Code
Interaction
Multi-agent
Environment
A simulator
Items from
Live play or votes
Refresh
Live
Run by
Anyone
In views
Agents, Kept fresh
Strongest model tested
Claude Sonnet 4 #113 on the capability index
Last model added
June 2025
Scores refreshed
November 20th, 2026
18
Entries in its paper, 17 models we track

DeepSeek-R1 leads with 39.6% among models released by June 2025 DeepSeek-R1 leads with 39.6% among models released by June 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. 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 June 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 DeepSeek-R1 May 2025 39.6%
  2. 2 o3-mini 38.8%
  3. 3 DeepSeek-V3 Mar 2025 35.9%
  4. 4 Qwen3-235B-A22B 34.9%
  5. 5 GPT-4.1 mini 34.5%
  6. 6 GPT-4.1 31.3%
  7. 7 codex-mini-2025-05-16 31.2%
  8. 8 Claude Sonnet 4 31.2%
  9. 9 Codestral 31.1%
  10. 10 Qwen2.5-72B 28.3%
  11. 11 Gemini 2.0 Flash Feb 2025 28.3%
  12. 12 Phi-4 27.6%
  13. 13 Gemini 2.5 Flash Jun 2025 25.1%
  14. 14 Llama 4 Maverick 21.5%
  15. 15 Claude 3.5 Sonnet (October 2024) 21%
  16. 16 Magistral Small 1.0 19.7%
  17. 17 Mercury-Coder 18.7%
  18. 18 Qwen2.5-Coder-32B-Instruct 13.4%

Source: The ProxyWar paper

#EntryScoreSettingDateSource
1 DeepSeek-R1 May 2025 39.6% May 2025 Official leaderboard
2 o3-mini 38.8% Jan 2025 Official leaderboard
3 DeepSeek-V3 Mar 2025 35.9% Mar 2025 Official leaderboard
4 Qwen3-235B-A22B 34.9% Apr 2025 Official leaderboard
5 GPT-4.1 mini 34.5% Apr 2025 Official leaderboard
6 GPT-4.1 31.3% Apr 2025 Official leaderboard
7 codex-mini-2025-05-16 31.2% May 2025 Official leaderboard
8 Claude Sonnet 4 31.2% May 2025 Official leaderboard
9 Codestral 31.1% Jan 2025 Official leaderboard
10 Qwen2.5-72B 28.3% Sep 2024 Official leaderboard
11 Gemini 2.0 Flash Feb 2025 28.3% Feb 2025 Official leaderboard
12 Phi-4 27.6% Dec 2024 Official leaderboard
13 Gemini 2.5 Flash Jun 2025 25.1% Jun 2025 Official leaderboard
14 Llama 4 Maverick 21.5% Apr 2025 Official leaderboard
15 Claude 3.5 Sonnet (October 2024) 21% Oct 2024 Official leaderboard
16 Magistral Small 1.0 19.7% Jun 2025 Official leaderboard
17 Mercury-Coder 18.7% Official leaderboard
18 Qwen2.5-Coder-32B-Instruct 13.4% Nov 2024 Official leaderboard
#ModelScoreSettingDateSource
1 DeepSeek-R1 May 2025 39.6% May 2025 Official leaderboard
2 o3-mini 38.8% Jan 2025 Official leaderboard
3 DeepSeek-V3 Mar 2025 35.9% Mar 2025 Official leaderboard
4 Qwen3-235B-A22B 34.9% Apr 2025 Official leaderboard
5 GPT-4.1 mini 34.5% Apr 2025 Official leaderboard
6 GPT-4.1 31.3% Apr 2025 Official leaderboard
7 codex-mini-2025-05-16 31.2% May 2025 Official leaderboard
8 Claude Sonnet 4 31.2% May 2025 Official leaderboard
9 Codestral 31.1% Jan 2025 Official leaderboard
10 Qwen2.5-72B 28.3% Sep 2024 Official leaderboard
11 Gemini 2.0 Flash Feb 2025 28.3% Feb 2025 Official leaderboard
12 Phi-4 27.6% Dec 2024 Official leaderboard
13 Gemini 2.5 Flash Jun 2025 25.1% Jun 2025 Official leaderboard
14 Llama 4 Maverick 21.5% Apr 2025 Official leaderboard
15 Claude 3.5 Sonnet (October 2024) 21% Oct 2024 Official leaderboard
16 Magistral Small 1.0 19.7% Jun 2025 Official leaderboard
17 Qwen2.5-Coder-32B-Instruct 13.4% Nov 2024 Official leaderboard

Rankings by part

We read 9 parts from the benchmark’s paper, each ranking models on one part of the test, such as Sudoku, Maze and 2048. Claude Sonnet 4 leads 4 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Sudoku

Average tournament rank. 17 models we track have a score here.

  1. 1 DeepSeek-R1 May 2025 1.40
  2. 2 Claude Sonnet 4 ↑6 3.00
  3. 3 DeepSeek-V3 Mar 2025 3.40
  4. 4 o3-mini ↓2 4.60
  5. 5 Qwen2.5-72B ↑5 6.00
  6. 6 Phi-4 ↑6 7.20
  7. 7 Qwen2.5-Coder-32B-Instruct ↑10 8.40
  8. 8 Magistral Small 1.0 ↑8 9.00
  9. 9 Claude 3.5 Sonnet (October 2024) ↑6 10.2
  10. 10 Gemini 2.0 Flash Feb 2025 ↑1 10.2

Number one over time

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

DeepSeek
Benchmark paper Not tested since Feb 2026
DeepSeek-R1 39.6%

Source: The ProxyWar paper

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
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) DeepSeek-R1 o3-mini Codestral Qwen2.5-72B 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) DeepSeek-R1 Codestral

Source: Official ProxyWar 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 Newest score check. The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (18 entries).
  • Pass Models scored 17 models we track, plus 1 other entry on its leaderboard.
  • Warning Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Sonnet 4 (#113).
  • Pass Headroom The best is 39.6%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 17 rows for models we track and 1 other entry 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