Benchmark Almanac

CrossWordBench

DeepSeek-R1 70.7% leads since March 2025
Not in model cards

Models solve generated crossword grids from text or images. The tasks vary the grid size and the share of letters already filled in.

The Markovian Thinker team independently ran CrossWordBench to test Delethink. The authors publish the generator and require registration for the hosted data.

Made by
HINT-lab and collaborators
Released
March 2025
Kind
Puzzles and chess
Level
School
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Text
Interaction
Single turn
Items from
Generated
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
Claude 3.7 Sonnet #119 on the capability index
Last model added
March 2025
Scores refreshed
November 20th, 2026
8
Models scored in its paper

DeepSeek-R1 leads with 70.7% among models released by March 2025

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline averages word correctness across the five puzzle sets. Where a model ran with several settings or harnesses, we keep 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 DeepSeek-R1 Text, zero-shot CoT 70.7%
  2. 2 o3-mini Text, zero-shot CoT, high reasoning effort 68%
  3. 3 QwQ-32B Text, zero-shot CoT 53%
  4. 4 Claude 3.7 Sonnet Text, zero-shot CoT, thinking off 50.6%
  5. 5 GPT-4o Nov 2024, Text, zero-shot CoT 46.1%
  6. 6 Pixtral Large Image, zero-shot CoT 29.1%
  7. 7 Qwen2.5-VL-72B Image, zero-shot CoT 29%
  8. 8 Gemini 2.0 Flash Feb 2025, Image, zero-shot CoT 26%

Source: The CrossWordBench paper

#ModelScoreSettingDateSource
1 DeepSeek-R1 70.7% Text, zero-shot CoT Jan 2025 Official leaderboard
2 o3-mini 68% Text, zero-shot CoT, high reasoning effort Jan 2025 Official leaderboard
3 QwQ-32B 53% Text, zero-shot CoT Mar 2025 Official leaderboard
4 Claude 3.7 Sonnet 50.6% Text, zero-shot CoT, thinking off Feb 2025 Official leaderboard
5 GPT-4o Nov 2024 46.1% Text, zero-shot CoT Nov 2024 Official leaderboard
6 Pixtral Large 29.1% Image, zero-shot CoT Nov 2024 Official leaderboard
7 Qwen2.5-VL-72B 29% Image, zero-shot CoT Jan 2025 Official leaderboard
8 Gemini 2.0 Flash Feb 2025 26% Image, zero-shot CoT Feb 2025 Official leaderboard

Rankings by part

We read 5 parts from the benchmark’s paper, each ranking models on one part of the test, such as English 7x7, English 14x14 and Chinese. DeepSeek-R1 leads 3 of them, and o3-mini leads the other 2. Arrows show how far a model moved from its overall rank among the models we track.

English 7x7

8 models we track have a score here.

  1. 1 DeepSeek-R1 64.6%
  2. 2 o3-mini 58.7%
  3. 3 Claude 3.7 Sonnet ↑1 48.2%
  4. 4 GPT-4o Nov 2024 ↑1 41%
  5. 5 QwQ-32B ↓2 34.7%
  6. 6 Pixtral Large 29.7%
  7. 7 Gemini 2.0 Flash Feb 2025 ↑1 27.7%
  8. 8 Qwen2.5-VL-72B ↓1 20.7%

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 Mar 2025
DeepSeek-R1 70.7%

Source: The CrossWordBench 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 GPT-4o Pixtral Large 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) DeepSeek-R1 GPT-4o

Source: Official CrossWordBench 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 20 months ago, in March 2025, for QwQ-32B. 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 (8 entries).
  • Pass Models scored 8 models we track.
  • Warning Newest score The newest came 20 months ago, in March 2025, for QwQ-32B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3.7 Sonnet (#119).
  • Pass Headroom The best is 70.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 10 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