Benchmark Almanac

GeoBench

Claude Fable 5 89% leads since Jun 2026 Dropped

Models geolocate GeoGuessr photos from community maps by giving coordinates and a country. Scored by country accuracy and a distance-based score.

Made by ccmdi Released Mar 2025 General visual questions All tasks public Last in a launch post Aug 2025 , dropped after it saturated Last model added Sep 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Short answer A number, word or short string matched to a key.
ccmdi/geobench 65 stars Leaderboard geobench.org
48
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab
2
System cards report it, from 1 lab

Claude Fable 5 leads with 89%

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.

48 models. Scroll sideways for the rest.

  1. 89.0 Claude Fable 5
  2. 88.0 Gemini 3 Flash
  3. 86.0 Gemini 2.5 Pro
  4. 86.0 Gemini 3.8 Flash
  5. 86.0 Muse Spark 1.3
  6. 85.0 Gemini 3.5 Flash
  7. 84.0 Gemini 2.5 Pro
  8. 84.0 Gemini 3 Pro
  9. 84.0 Gemini 3.1 Pro
  10. 84.0 Claude Fable 5.1
  11. 83.0 Gemini 3.1 Flash-Lite
  12. 81.0 Gemini 2.5 Pro
  13. 81.0 GPT-5
  14. 81.0 Claude Opus 4.8
  15. 80.0 o1
  16. 77.5 Gemini 2.0 Pro
  17. 77.0 Gemini 2.0 Flash
  18. 76.0 Gemini 1.5 Flash
  19. 76.0 Gemini 2.5 Flash
  20. 75.0 Claude Opus 4.5
  21. 74.0 o3
  22. 74.0 Claude Sonnet 5
  23. 73.0 Gemini 2.5 Flash
  24. 73.0 Claude Sonnet 4.6
  25. 73.0 GPT-5.6 Sol
  26. 72.0 GPT-4.1
  27. 71.0 GPT-4o
  28. 71.0 GLM-4.6V
  29. 70.0 Gemini 2.0 Flash Thinking
  30. 68.0 Claude 3.7 Sonnet
  31. 64.0 GPT-4o mini
  32. 64.0 o4-mini
  33. 62.0 Claude 3.5 Sonnet
  34. 62.0 Qwen2.5-VL-72B
  35. 61.0 Grok 4.7
  36. 60.0 Claude Sonnet 4.5
  37. 59.0 Qwen3-VL-235B-A22B
  38. 56.0 Mistral Medium 3.5
  39. 56.0 Grok 4.3
  40. 52.0 Llama 3.2 90B
  41. 52.0 Gemma 3 27B
  42. 52.0 Llama 4 Maverick
  43. 49.0 Claude Opus 4
  44. 45.0 Grok 4
  45. 38.0 Qwen 3.8 27B
  46. 37.0 Pixtral 12B
  47. 37.0 Claude Sonnet 4
  48. 34.0 Claude 3.5 Haiku
  1. 1 Claude Fable 5 89%
  2. 2 Gemini 3 Flash 88%
  3. 3 Gemini 2.5 Pro May 2025 86%
  4. 4 Gemini 3.8 Flash 86%
  5. 5 Muse Spark 1.3 86%
  6. 6 Gemini 3.5 Flash 85%
  7. 7 Gemini 2.5 Pro Mar 2025, with search 84%
  8. 8 Gemini 3 Pro 84%
  9. 9 Gemini 3.1 Pro 84%
  10. 10 Claude Fable 5.1 84%
  11. 11 Gemini 3.1 Flash-Lite 83%
  12. 12 Gemini 2.5 Pro Jun 2025 81%
  13. 13 GPT-5 81%
  14. 14 Claude Opus 4.8 thinking 81%
  15. 15 o1 80%
  16. 16 Gemini 2.0 Pro 77.5%
  17. 17 Gemini 2.0 Flash Feb 2025 77%
  18. 18 Gemini 1.5 Flash Sep 2024 76%
  19. 19 Gemini 2.5 Flash May 2025 76%
  20. 20 Claude Opus 4.5 thinking 75%
  21. 21 o3 74%
  22. 22 Claude Sonnet 5 74%
  23. 23 Gemini 2.5 Flash Apr 2025 73%
  24. 24 Claude Sonnet 4.6 thinking 73%
  25. 25 GPT-5.6 Sol high 73%
  26. 26 GPT-4.1 72%
  27. 27 GPT-4o Jan 2025 71%
  28. 28 GLM-4.6V 71%
  29. 29 Gemini 2.0 Flash Thinking Jan 2025 70%
  30. 30 Claude 3.7 Sonnet 68%
  31. 31 GPT-4o mini 64%
  32. 32 o4-mini high 64%
  33. 33 Claude 3.5 Sonnet 62%
  34. 34 Qwen2.5-VL-72B 62%
  35. 35 Grok 4.7 61%
  36. 36 Claude Sonnet 4.5 thinking 60%
  37. 37 Qwen3-VL-235B-A22B 59%
  38. 38 Mistral Medium 3.5 56%
  39. 39 Grok 4.3 56%
  40. 40 Llama 3.2 90B 52%
  41. 41 Gemma 3 27B 52%
  42. 42 Llama 4 Maverick 52%
  43. 43 Claude Opus 4 thinking 49%
  44. 44 Grok 4 45%
  45. 45 Qwen 3.8 27B 38%
  46. 46 Pixtral 12B 37%
  47. 47 Claude Sonnet 4 37%
  48. 48 Claude 3.5 Haiku 34%
#ModelScoreSettingDateSource
1 Claude Fable 5 89% Jun 2026 Official leaderboard
2 Gemini 3 Flash 88% Dec 2025 Official leaderboard
3 Gemini 2.5 Pro May 2025 86% May 2025 Official leaderboard
4 Gemini 3.8 Flash 86% Sep 2026 Official leaderboard
5 Muse Spark 1.3 86% Sep 2026 Official leaderboard
6 Gemini 3.5 Flash 85% May 2026 Official leaderboard
7 Gemini 2.5 Pro Mar 2025 84% with search Mar 2025 Official leaderboard
8 Gemini 3 Pro 84% Nov 2025 Official leaderboard
9 Gemini 3.1 Pro 84% Feb 2026 Official leaderboard
10 Claude Fable 5.1 84% Sep 2026 Official leaderboard
11 Gemini 3.1 Flash-Lite 83% Mar 2026 Official leaderboard
12 Gemini 2.5 Pro Jun 2025 81% Jun 2025 Official leaderboard
13 GPT-5 81% Aug 2025 Official leaderboard
14 Claude Opus 4.8 81% thinking May 2026 Official leaderboard
15 o1 80% Dec 2024 Official leaderboard
16 Gemini 2.0 Pro 77.5% Feb 2025 Official leaderboard
17 Gemini 2.0 Flash Feb 2025 77% Feb 2025 Official leaderboard
18 Gemini 1.5 Flash Sep 2024 76% Aug 2024 Official leaderboard
19 Gemini 2.5 Flash May 2025 76% May 2025 Official leaderboard
20 Claude Opus 4.5 75% thinking Nov 2025 Official leaderboard
21 o3 74% Apr 2025 Official leaderboard
22 Claude Sonnet 5 74% Jun 2026 Official leaderboard
23 Gemini 2.5 Flash Apr 2025 73% Apr 2025 Official leaderboard
24 Claude Sonnet 4.6 73% thinking Feb 2026 Official leaderboard
25 GPT-5.6 Sol 73% high Jul 2026 Official leaderboard
26 GPT-4.1 72% Apr 2025 Official leaderboard
27 GPT-4o Jan 2025 71% Jan 2025 Official leaderboard
28 GLM-4.6V 71% Dec 2025 Official leaderboard
29 Gemini 2.0 Flash Thinking Jan 2025 70% Jan 2025 Official leaderboard
30 Claude 3.7 Sonnet 68% Feb 2025 Official leaderboard
31 GPT-4o mini 64% Jul 2024 Official leaderboard
32 o4-mini 64% high Apr 2025 Official leaderboard
33 Claude 3.5 Sonnet 62% Jun 2024 Official leaderboard
34 Qwen2.5-VL-72B 62% Jan 2025 Official leaderboard
35 Grok 4.7 61% Sep 2026 Official leaderboard
36 Claude Sonnet 4.5 60% thinking Sep 2025 Official leaderboard
37 Qwen3-VL-235B-A22B 59% Sep 2025 Official leaderboard
38 Mistral Medium 3.5 56% Apr 2026 Official leaderboard
39 Grok 4.3 56% Apr 2026 Official leaderboard
40 Llama 3.2 90B 52% Sep 2024 Official leaderboard
41 Gemma 3 27B 52% Mar 2025 Official leaderboard
42 Llama 4 Maverick 52% Apr 2025 Official leaderboard
43 Claude Opus 4 49% thinking May 2025 Official leaderboard
44 Grok 4 45% Jul 2025 Official leaderboard
45 Qwen 3.8 27B 38% Aug 2026 Official leaderboard
46 Pixtral 12B 37% Sep 2024 Official leaderboard
47 Claude Sonnet 4 37% May 2025 Official leaderboard
48 Claude 3.5 Haiku 34% Oct 2024 Official leaderboard

Epoch AI has scores for 27 models. Labs report scores for 2 models in their cards. We plot every source in the charts below.

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Gemini 2.5 Pro
with search
86%
Gemini 2.5 Pro
with search
84%
o3
high
74%
Claude 3.7 Sonnet
thinking
68%
o4-mini
high
64%
Claude Sonnet 4
thinking
37%
ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5 89%
Gemini 3 Flash 88%
Gemini 2.5 Pro May 2025 86%
Gemini 3.8 Flash 86%
Muse Spark 1.3 86%
Gemini 3.5 Flash 85%
Gemini 2.5 Pro Mar 2025 with search 84%
Gemini 3 Pro 84%
Gemini 3.1 Pro 84%
Claude Fable 5.1 84%
Gemini 3.1 Flash-Lite 83%
Gemini 2.5 Pro May 2025 with search 81%
Gemini 2.5 Pro Jun 2025 81%
GPT-5 81%
Claude Opus 4.8 thinking 81%
o1 80%
Gemini 2.5 Pro Mar 2025 78%
Gemini 2.0 Pro 77.5%
Gemini 2.0 Flash Feb 2025 77%
Gemini 1.5 Flash Sep 2024 76%
Gemini 2.5 Flash May 2025 76%
Claude Opus 4.5 thinking 75%
o3 74%
Claude Sonnet 5 74%
Gemini 2.5 Flash Apr 2025 73%
Claude Sonnet 4.6 thinking 73%
GPT-5.6 Sol high 73%
GPT-4.1 72%
GPT-4o Jan 2025 71%
GLM-4.6V 71%
Gemini 2.0 Flash Thinking Jan 2025 70%
Claude 3.7 Sonnet 68%
Claude 3.7 Sonnet thinking 65%
GPT-4o mini 64%
o4-mini high 64%
o4-mini 64%
Claude 3.5 Sonnet 62%
Qwen2.5-VL-72B 62%
Grok 4.7 61%
o3 high 60%
Claude Sonnet 4.5 thinking 60%
Qwen3-VL-235B-A22B 59%
Mistral Medium 3.5 56%
Grok 4.3 56%
Llama 3.2 90B 52%
Gemma 3 27B 52%
Llama 4 Maverick 52%
Claude Opus 4 thinking 49%
Grok 4 45%
Qwen 3.8 27B 38%
Pixtral 12B 37%
Claude Sonnet 4 37%
Claude 3.5 Haiku 34%
Claude Sonnet 4 thinking 32%

5 models have held number one since its release in Mar 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard, Epoch AI’s runs and what labs claim in their cards.

OpenAI Google DeepMind Anthropic Z.ai
Own leaderboard Passed 85% in May 2025, by Gemini 2.5 Pro
Claude Fable 5 89%
Epoch AI runs Passed 85% in May 2025, by Gemini 2.5 Pro
Gemini 3 Flash 88%
Lab claims Not tested since Aug 2025
GLM-4.5V 79.7%

The best score passed 85% in May 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 Run or collected by Epoch AI Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Claude Fable 5 Gemini 3 Flash Gemini 2.5 Pro o1 GPT-4o mini 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Claude Fable 5 o1

Labs dropped it from launch posts once the best score passed 85%

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.

Z.ai
DateLaunch postScoreAs the post writes it
11 Aug 2025 GLM-4.5V 79.7 GeoBench

2 system cards report it, from 1 lab

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
13 Aug 2025 GLM-4.5V technical report 79.7 GeoBench, normalized to a percentage scale, GLM-4.5V thinking (Table 2) Capabilities
1 Jul 2025 GLM-4.1V-9B-Thinking technical report 76 GeoBench, normalized to a percentage scale, GLM-4.1V-9B-Thinking thinking (Table 2) 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%