Benchmark Almanac

Math Kangaroo 2025 (MathArena)

GPT-5.5 94.9% leads since Apr 2026 Fading

Vision language models answer 168 multiple-choice problems from the 2025 Math Kangaroo, given as screenshots across six grade groups. Many problems need reading a diagram.

Models score lower on problems for grades 1 to 6 than on those for grades 7 to 12. MathArena ran 23 models and later marked the set deprecated.

Made by MathArena (ETH Zurich SRI Lab and INSAIT) Released Oct 2025 Visual math and diagrams All tasks public Last in a launch post Feb 2026 , dropped after it saturated Last model added Jul 2026 Scores refreshed 2 Nov 2026
Level
School Students, from grade school to high school
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Multiple choice
eth-sri/matharena 294 stars Dataset Leaderboard matharena.ai
22
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab
1
System card reports it, from 1 lab

GPT-5.5 leads with 94.9% among models released by Jul 2026

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in Jul 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

22 models. Scroll sideways for the rest.

  1. 94.9 GPT-5.5
  2. 92.5 GPT-5.4
  3. 91.3 Kimi K3
  4. 89.9 Gemini 3.5 Flash
  5. 89.4 Gemini 3.1 Pro
  6. 87.8 Gemini 3.6 Flash
  7. 86.5 GPT-5.2
  8. 85.8 Gemini 3 Flash
  9. 84.2 Gemini 3 Pro
  10. 81.6 Claude Opus 4.8
  11. 80.6 Kimi K2.5
  12. 78.8 GPT-5
  13. 78.2 GPT-5 mini
  14. 77.2 Gemini 2.5 Pro
  15. 76.9 GPT-5.1
  16. 75.8 Claude Sonnet 4.5
  17. 72.5 Qwen3-VL-235B-A22B-Instruct
  18. 72.3 Claude Opus 4.6
  19. 70.0 Grok 4
  20. 69.0 Grok 4.1 Fast
  21. 67.6 GLM-4.5V
  22. 66.8 Grok 4 Fast
  1. 1 GPT-5.5 xhigh 94.9%
  2. 2 GPT-5.4 xhigh 92.5%
  3. 3 Kimi K3 thinking 91.3%
  4. 4 Gemini 3.5 Flash 89.9%
  5. 5 Gemini 3.1 Pro 89.4%
  6. 6 Gemini 3.6 Flash 87.8%
  7. 7 GPT-5.2 high 86.5%
  8. 8 Gemini 3 Flash 85.8%
  9. 9 Gemini 3 Pro 84.2%
  10. 10 Claude Opus 4.8 max 81.6%
  11. 11 Kimi K2.5 thinking 80.6%
  12. 12 GPT-5 high 78.8%
  13. 13 GPT-5 mini high 78.2%
  14. 14 Gemini 2.5 Pro Jun 2025 77.2%
  15. 15 GPT-5.1 high 76.9%
  16. 16 Claude Sonnet 4.5 thinking 75.8%
  17. 17 Qwen3-VL-235B-A22B-Instruct 72.5%
  18. 18 Claude Opus 4.6 high 72.3%
  19. 19 Grok 4 70%
  20. 20 Grok 4.1 Fast 69%
  21. 21 GLM-4.5V 67.6%
  22. 22 Grok 4 Fast 66.8%
#ModelScoreSettingDateSource
1 GPT-5.5 94.9% xhigh Apr 2026 Official leaderboard
2 GPT-5.4 92.5% xhigh Mar 2026 Official leaderboard
3 Kimi K3 91.3% thinking Jul 2026 Official leaderboard
4 Gemini 3.5 Flash 89.9% May 2026 Official leaderboard
5 Gemini 3.1 Pro 89.4% Feb 2026 Official leaderboard
6 Gemini 3.6 Flash 87.8% Jul 2026 Official leaderboard
7 GPT-5.2 86.5% high Dec 2025 Official leaderboard
8 Gemini 3 Flash 85.8% Dec 2025 Official leaderboard
9 Gemini 3 Pro 84.2% Nov 2025 Official leaderboard
10 Claude Opus 4.8 81.6% max May 2026 Official leaderboard
11 Kimi K2.5 80.6% thinking Jan 2026 Official leaderboard
12 GPT-5 78.8% high Aug 2025 Official leaderboard
13 GPT-5 mini 78.2% high Aug 2025 Official leaderboard
14 Gemini 2.5 Pro Jun 2025 77.2% Jun 2025 Official leaderboard
15 GPT-5.1 76.9% high Nov 2025 Official leaderboard
16 Claude Sonnet 4.5 75.8% thinking Sep 2025 Official leaderboard
17 Qwen3-VL-235B-A22B-Instruct 72.5% Sep 2025 Official leaderboard
18 Claude Opus 4.6 72.3% high Feb 2026 Official leaderboard
19 Grok 4 70% Jul 2025 Official leaderboard
20 Grok 4.1 Fast 69% Nov 2025 Official leaderboard
21 GLM-4.5V 67.6% Aug 2025 Official leaderboard
22 Grok 4 Fast 66.8% Sep 2025 Official leaderboard

Labs report scores for 1 model in their cards. We plot every source in the charts below.

The best score, 94.9%, costs $0.12 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.

OpenAI xAI Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-5.5, xhigh Gemini 3.5 Flash Gemini 3 Flash GPT-5 mini, high Grok 4.1 Fast 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-5.5, xhigh Gemini 3.6 Flash
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.5 xhigh 94.9% $0.12 3,883
GPT-5.4 xhigh 92.5% $0.09 5,580
Kimi K3 thinking 91.3% $0.10 6,511
Gemini 3.5 Flash 89.9% $0.06 6,490
Gemini 3.1 Pro 89.4% $0.15 12,821
Gemini 3.6 Flash 87.8% $0.04 5,466
GPT-5.2 high 86.5% $0.06 4,133
Gemini 3 Flash 85.8% $0.03 9,336
Gemini 3 Pro 84.2% $0.11 9,352
Claude Opus 4.8 max 81.6% $0.54 21,731
Kimi K2.5 thinking 80.6% $0.03 9,573
GPT-5 high 78.8% $0.07 7,243
GPT-5 mini high 78.2% $0.01 5,047
Gemini 2.5 Pro Jun 2025 77.2% $0.11 11,113
GPT-5.1 high 76.9% $0.06 5,653
Claude Sonnet 4.5 thinking 75.8% $0.09 5,565
Qwen3-VL-235B-A22B-Instruct 72.5% $0.01 2,713
Claude Opus 4.6 high 72.3% $0.23 8,946
Grok 4 70% $0.17 11,241
Grok 4.1 Fast 69% $0.004 7,716
GLM-4.5V 67.6% $0.005 2,167
Grok 4 Fast 66.8% $0.003 5,918

6 models have held number one since its release in Oct 2025

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

OpenAI Google DeepMind
Own leaderboard Passed 85% in Dec 2025, by GPT-5.2
GPT-5.5 94.9%

The best score passed 85% in Dec 2025

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

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.5 Seed2.0 Pro Gemini 3 Pro GPT-5 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.5 GPT-5.2

Labs headline it less often now, last in Feb 2026

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

ByteDance Seed
DateLaunch postScoreAs the post writes it
14 Feb 2026 Seed2.0 Pro 90.5 MathKangaroo

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
14 Feb 2026 Seed2.0 Pro model card 90.5 MathKangaroo, mean over 2025 competitions (Table 8) Capabilities

More in Vision

All 234 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. Kimi K3.1 Pro 84.1%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%
  • DocVQA In use Models answer 50,000 questions about 12,000 scanned document images by reading their text and layout. Scoring is ANLS, a string similarity to reference answers. Seed1.5-VL 96.9%