Benchmark Almanac

IOI (Vals)

Claude Opus 5 91.7% leads since July 2026
Not in model cards

Agents solve International Olympiad in Informatics problems from 2024 and 2025 in C++. A submission tool modeled on the contest grades up to 50 tries per problem by subtask.

Vals archived v1 as saturated after August 9th, 2026, with Claude Opus 5 first at 91.67%. Not the same as the almanac's single-year IOI 2025 record.

Made by
Vals AI
Released
August 2025
Kind
Competitive programming
Level
Competition
Scored
Checked automatically
Answers
Code
Tasks
All public
Strongest model tested
Claude Opus 5 #10 on the capability index
Last model added
August 2026
Scores refreshed
November 12th, 2026
55
Models scored on its own leaderboard

Claude Opus 5 leads with 91.7% among models released by August 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 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 August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

55 models. Scroll sideways for the rest.

  1. 91.7 Claude Opus 5
  2. 86.7 GPT-5.6 Sol
  3. 73.0 Qwen 3.8 Max
  4. 72.9 GPT-5.6 Luna
  5. 72.3 Claude Fable 5
  6. 67.8 GPT-5.4
  7. 65.3 GPT-5.6 Terra
  8. 54.8 GPT-5.2
  9. 49.5 Muse Spark 1.2
  10. 47.1 Claude Opus 4.7
  11. 46.8 Qwen3.7-Max
  12. 43.8 GPT-5.3 Codex
  13. 39.1 Gemini 3 Flash
  14. 38.8 Gemini 3 Pro
  15. 35.8 DeepSeek-V4-Pro
  16. 30.2 Grok 4.20
  17. 26.2 Grok 4
  18. 26.2 Gemini 3.5 Flash-Lite
  19. 23.6 Claude Opus 4.5
  20. 22.0 GLM-5
  21. 21.5 GPT-5.1
  22. 21.4 GPT-5.1-Codex-Max
  23. 20.0 GPT-5
  24. 18.3 Claude Sonnet 4.5
  25. 17.7 Kimi K2.5
  26. 17.1 Gemini 2.5 Pro
  27. 15.7 Qwen3-Max
  28. 15.3 Grok 4.3
  29. 15.3 GPT-5.4 Nano
  30. 14.4 DeepSeek-V3.2
  31. 12.5 Claude Opus 4.1
  32. 11.5 Grok 4 Fast
  33. 9.75 GPT-5-Codex
  34. 7.67 Grok 4.1 Fast
  35. 7.58 GLM-4.7
  36. 6.75 GPT-5 mini
  37. 6.67 MiniMax-M2.5
  38. 6.50 Claude Sonnet 4
  39. 6.42 GPT-5.4 Mini
  40. 6.17 Claude Haiku 4.5
  41. 4.92 MiniMax-M2.7
  42. 4.83 o4-mini
  43. 4.33 grok-code-fast-1
  44. 4.33 GLM-4.6
  45. 4.00 Mistral Large 3
  46. 3.67 GPT-5.1-Codex
  47. 2.92 GLM-4.5
  48. 2.61 Gemini 2.5 Flash
  49. 2.50 Devstral Small 2
  50. 2.33 MiniMax-M2.1
  51. 1.67 DeepSeek-V3
  52. 1.25 Kimi K2
  53. 0.67 Magistral Medium 1.2
  54. 0.67 Devstral 2
  55. 0.0 Qwen3-235B-A22B
  1. 1 Claude Opus 5 91.7%
  2. 2 GPT-5.6 Sol 86.7%
  3. 3 Qwen 3.8 Max 73%
  4. 4 GPT-5.6 Luna 72.9%
  5. 5 Claude Fable 5 72.3%
  6. 6 GPT-5.4 xhigh 67.8%
  7. 7 GPT-5.6 Terra 65.3%
  8. 8 GPT-5.2 54.8%
  9. 9 Muse Spark 1.2 49.5%
  10. 10 Claude Opus 4.7 47.1%
  11. 11 Qwen3.7-Max 46.8%
  12. 12 GPT-5.3 Codex 43.8%
  13. 13 Gemini 3 Flash 39.1%
  14. 14 Gemini 3 Pro 38.8%
  15. 15 DeepSeek-V4-Pro 35.8%
  16. 16 Grok 4.20 reasoning 30.2%
  17. 17 Grok 4 26.2%
  18. 18 Gemini 3.5 Flash-Lite 26.2%
  19. 19 Claude Opus 4.5 non-thinking 23.6%
  20. 20 GLM-5 thinking 22%
  21. 21 GPT-5.1 21.5%
  22. 22 GPT-5.1-Codex-Max 21.4%
  23. 23 GPT-5 20%
  24. 24 Claude Sonnet 4.5 thinking 18.3%
  25. 25 Kimi K2.5 thinking 17.7%
  26. 26 Gemini 2.5 Pro Jun 2025 17.1%
  27. 27 Qwen3-Max 15.7%
  28. 28 Grok 4.3 15.3%
  29. 29 GPT-5.4 Nano 15.3%
  30. 30 DeepSeek-V3.2 non-thinking 14.4%
  31. 31 Claude Opus 4.1 non-thinking 12.5%
  32. 32 Grok 4 Fast reasoning 11.5%
  33. 33 GPT-5-Codex 9.75%
  34. 34 Grok 4.1 Fast non-reasoning 7.67%
  35. 35 GLM-4.7 7.58%
  36. 36 GPT-5 mini 6.75%
  37. 37 MiniMax-M2.5 6.67%
  38. 38 Claude Sonnet 4 non-thinking 6.50%
  39. 39 GPT-5.4 Mini 6.42%
  40. 40 Claude Haiku 4.5 thinking 6.17%
  41. 41 MiniMax-M2.7 4.92%
  42. 42 o4-mini 4.83%
  43. 43 grok-code-fast-1 4.33%
  44. 44 GLM-4.6 4.33%
  45. 45 Mistral Large 3 4.00%
  46. 46 GPT-5.1-Codex 3.67%
  47. 47 GLM-4.5 2.92%
  48. 48 Gemini 2.5 Flash Jun 2025, non-thinking 2.61%
  49. 49 Devstral Small 2 2.50%
  50. 50 MiniMax-M2.1 2.33%
  51. 51 DeepSeek-V3 Mar 2025 1.67%
  52. 52 Kimi K2 Jul 2025 1.25%
  53. 53 Magistral Medium 1.2 0.67%
  54. 54 Devstral 2 0.67%
  55. 55 Qwen3-235B-A22B 0%

Source: Official IOI (Vals) leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5 91.7% Jul 2026 Official leaderboard
2 GPT-5.6 Sol 86.7% Jul 2026 Official leaderboard
3 Qwen 3.8 Max 73% Aug 2026 Official leaderboard
4 GPT-5.6 Luna 72.9% Jul 2026 Official leaderboard
5 Claude Fable 5 72.3% Jun 2026 Official leaderboard
6 GPT-5.4 67.8% xhigh Mar 2026 Official leaderboard
7 GPT-5.6 Terra 65.3% Jul 2026 Official leaderboard
8 GPT-5.2 54.8% Dec 2025 Official leaderboard
9 Muse Spark 1.2 49.5% Aug 2026 Official leaderboard
10 Claude Opus 4.7 47.1% Apr 2026 Official leaderboard
11 Qwen3.7-Max 46.8% May 2026 Official leaderboard
12 GPT-5.3 Codex 43.8% Feb 2026 Official leaderboard
13 Gemini 3 Flash 39.1% Dec 2025 Official leaderboard
14 Gemini 3 Pro 38.8% Nov 2025 Official leaderboard
15 DeepSeek-V4-Pro 35.8% Apr 2026 Official leaderboard
16 Grok 4.20 30.2% reasoning Feb 2026 Official leaderboard
17 Grok 4 26.2% Jul 2025 Official leaderboard
18 Gemini 3.5 Flash-Lite 26.2% Jul 2026 Official leaderboard
19 Claude Opus 4.5 23.6% non-thinking Nov 2025 Official leaderboard
20 GLM-5 22% thinking Feb 2026 Official leaderboard
21 GPT-5.1 21.5% Nov 2025 Official leaderboard
22 GPT-5.1-Codex-Max 21.4% Nov 2025 Official leaderboard
23 GPT-5 20% Aug 2025 Official leaderboard
24 Claude Sonnet 4.5 18.3% thinking Sep 2025 Official leaderboard
25 Kimi K2.5 17.7% thinking Jan 2026 Official leaderboard
26 Gemini 2.5 Pro Jun 2025 17.1% Jun 2025 Official leaderboard
27 Qwen3-Max 15.7% Sep 2025 Official leaderboard
28 Grok 4.3 15.3% Apr 2026 Official leaderboard
29 GPT-5.4 Nano 15.3% Mar 2026 Official leaderboard
30 DeepSeek-V3.2 14.4% non-thinking Dec 2025 Official leaderboard
31 Claude Opus 4.1 12.5% non-thinking Aug 2025 Official leaderboard
32 Grok 4 Fast 11.5% reasoning Sep 2025 Official leaderboard
33 GPT-5-Codex 9.75% Sep 2025 Official leaderboard
34 Grok 4.1 Fast 7.67% non-reasoning Nov 2025 Official leaderboard
35 GLM-4.7 7.58% Dec 2025 Official leaderboard
36 GPT-5 mini 6.75% Aug 2025 Official leaderboard
37 MiniMax-M2.5 6.67% Feb 2026 Official leaderboard
38 Claude Sonnet 4 6.50% non-thinking May 2025 Official leaderboard
39 GPT-5.4 Mini 6.42% Mar 2026 Official leaderboard
40 Claude Haiku 4.5 6.17% thinking Oct 2025 Official leaderboard
41 MiniMax-M2.7 4.92% Mar 2026 Official leaderboard
42 o4-mini 4.83% Apr 2025 Official leaderboard
43 grok-code-fast-1 4.33% Aug 2025 Official leaderboard
44 GLM-4.6 4.33% Sep 2025 Official leaderboard
45 Mistral Large 3 4.00% Dec 2025 Official leaderboard
46 GPT-5.1-Codex 3.67% Nov 2025 Official leaderboard
47 GLM-4.5 2.92% Aug 2025 Official leaderboard
48 Gemini 2.5 Flash Jun 2025 2.61% non-thinking Jun 2025 Official leaderboard
49 Devstral Small 2 2.50% Dec 2025 Official leaderboard
50 MiniMax-M2.1 2.33% Dec 2025 Official leaderboard
51 DeepSeek-V3 Mar 2025 1.67% Mar 2025 Official leaderboard
52 Kimi K2 Jul 2025 1.25% Jul 2025 Official leaderboard
53 Magistral Medium 1.2 0.67% Sep 2025 Official leaderboard
54 Devstral 2 0.67% Dec 2025 Official leaderboard
55 Qwen3-235B-A22B 0% Apr 2025 Official leaderboard

Accuracy against latency

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% 100% 1001,00010K100K Latency per test (s), log scale, lower is better Accuracy o4-mini: Accuracy 4.83%, latency per test 480 s Qwen3-235B-A22B: Accuracy 0%, latency per test 353 s Claude Sonnet 4 (non-thinking): Accuracy 6.5%, latency per test 3,253 s Claude Sonnet 4 (thinking): Accuracy 4.58%, latency per test 2,216 s Gemini 2.5 Pro: Accuracy 17.1%, latency per test 2,965 s Grok 4: Accuracy 26.2%, latency per test 4,266 s Kimi K2: Accuracy 1.25%, latency per test 1,551 s GLM-4.5: Accuracy 2.92%, latency per test 2,336 s Claude Opus 4.1 (non-thinking): Accuracy 12.5%, latency per test 6,125 s GPT-5: Accuracy 20%, latency per test 2,608 s grok-code-fast-1: Accuracy 4.33%, latency per test 1,084 s GPT-5-Codex: Accuracy 9.75%, latency per test 1,543 s Magistral Medium 1.2: Accuracy 0.67%, latency per test 2,212 s Grok 4 Fast (reasoning): Accuracy 11.5%, latency per test 1,409 s Grok 4 Fast (non-reasoning): Accuracy 3.83%, latency per test 578 s Qwen3-Max: Accuracy 15.7%, latency per test 2,045 s Qwen3-Max: Accuracy 13.8%, latency per test 2,953 s Qwen3-Max: Accuracy 7.75%, latency per test 3,372 s Claude Sonnet 4.5 (thinking): Accuracy 18.3%, latency per test 3,315 s GLM-4.6: Accuracy 4.33%, latency per test 7,106 s Claude Haiku 4.5 (thinking): Accuracy 6.17%, latency per test 7,415 s GPT-5.1-Codex: Accuracy 3.67%, latency per test 2,227 s GPT-5.1: Accuracy 21.5%, latency per test 1,750 s GPT-5.1-Codex-Max: Accuracy 21.4%, latency per test 957 s Grok 4.1 Fast (non-reasoning): Accuracy 7.67%, latency per test 7,675 s Grok 4.1 Fast (reasoning): Accuracy 3.08%, latency per test 1,533 s Claude Opus 4.5 (non-thinking): Accuracy 23.6%, latency per test 7,682 s Claude Opus 4.5 (thinking): Accuracy 20.3%, latency per test 8,212 s DeepSeek-V3.2 (non-thinking): Accuracy 14.4%, latency per test 2,385 s DeepSeek-V3.2 (thinking): Accuracy 10.7%, latency per test 1,559 s Mistral Large 3: Accuracy 4%, latency per test 6,789 s Devstral 2: Accuracy 0.67%, latency per test 8,045 s Devstral Small 2: Accuracy 2.5%, latency per test 7,467 s GPT-5.2: Accuracy 54.8%, latency per test 6,740 s Gemini 3 Flash: Accuracy 39.1%, latency per test 1,847 s GLM-4.7: Accuracy 7.58%, latency per test 5,317 s MiniMax-M2.1: Accuracy 2.33%, latency per test 5,210 s Kimi K2.5 (thinking): Accuracy 17.7%, latency per test 3,828 s GPT-5.3 Codex: Accuracy 43.8%, latency per test 3,574 s GLM-5 (thinking): Accuracy 22%, latency per test 6,807 s MiniMax-M2.5: Accuracy 6.67%, latency per test 23K s GPT-5.4 (xhigh): Accuracy 67.8%, latency per test 2,661 s GPT-5.4 Mini: Accuracy 6.42%, latency per test 1,897 s GPT-5.4 Nano: Accuracy 15.3%, latency per test 1,171 s MiniMax-M2.7: Accuracy 4.92%, latency per test 3,243 s Claude Opus 4.7: Accuracy 47.1%, latency per test 3,082 s DeepSeek-V4-Pro: Accuracy 35.8%, latency per test 5,374 s Qwen3.7-Max: Accuracy 46.8%, latency per test 4,380 s Claude Fable 5: Accuracy 72.3%, latency per test 3,319 s GPT-5.6 Luna: Accuracy 72.9%, latency per test 1,671 s GPT-5.6 Terra: Accuracy 65.3%, latency per test 1,331 s Qwen 3.8 Max: Accuracy 73%, latency per test 3,946 s Muse Spark 1.2: Accuracy 49.5%, latency per test 4,772 s DeepSeek-V3: Accuracy 1.67%, latency per test 171 s Gemini 2.5 Flash (non-thinking): Accuracy 2.61%, latency per test 350 s GPT-5 mini: Accuracy 6.75%, latency per test 390 s Gemini 3 Pro: Accuracy 38.8%, latency per test 900 s Grok 4.20 (reasoning): Accuracy 30.2%, latency per test 897 s Grok 4.3: Accuracy 15.3%, latency per test 629 s GPT-5.6 Sol: Accuracy 86.7%, latency per test 1,311 s Gemini 3.5 Flash-Lite: Accuracy 26.2%, latency per test 882 s Claude Opus 5: Accuracy 91.7%, latency per test 1,693 s Claude Opus 5GPT-5.6 SolGemini 3 ProGrok 4.20 (reasoning)Gemini 3.5 Flash-LiteGrok 4.3GPT-5 miniGemini 2.5 Flash (non-thinking)DeepSeek-V3 0% 20% 40% 60% 80% 100% 1001,00010K100K Latency per test (s), log scale, lower is better Accuracy o4-mini: Accuracy 4.83%, latency per test 480 s Qwen3-235B-A22B: Accuracy 0%, latency per test 353 s Claude Sonnet 4 (non-thinking): Accuracy 6.5%, latency per test 3,253 s Claude Sonnet 4 (thinking): Accuracy 4.58%, latency per test 2,216 s Gemini 2.5 Pro: Accuracy 17.1%, latency per test 2,965 s Grok 4: Accuracy 26.2%, latency per test 4,266 s Kimi K2: Accuracy 1.25%, latency per test 1,551 s GLM-4.5: Accuracy 2.92%, latency per test 2,336 s Claude Opus 4.1 (non-thinking): Accuracy 12.5%, latency per test 6,125 s GPT-5: Accuracy 20%, latency per test 2,608 s grok-code-fast-1: Accuracy 4.33%, latency per test 1,084 s GPT-5-Codex: Accuracy 9.75%, latency per test 1,543 s Magistral Medium 1.2: Accuracy 0.67%, latency per test 2,212 s Grok 4 Fast (reasoning): Accuracy 11.5%, latency per test 1,409 s Grok 4 Fast (non-reasoning): Accuracy 3.83%, latency per test 578 s Qwen3-Max: Accuracy 15.7%, latency per test 2,045 s Qwen3-Max: Accuracy 13.8%, latency per test 2,953 s Qwen3-Max: Accuracy 7.75%, latency per test 3,372 s Claude Sonnet 4.5 (thinking): Accuracy 18.3%, latency per test 3,315 s GLM-4.6: Accuracy 4.33%, latency per test 7,106 s Claude Haiku 4.5 (thinking): Accuracy 6.17%, latency per test 7,415 s GPT-5.1-Codex: Accuracy 3.67%, latency per test 2,227 s GPT-5.1: Accuracy 21.5%, latency per test 1,750 s GPT-5.1-Codex-Max: Accuracy 21.4%, latency per test 957 s Grok 4.1 Fast (non-reasoning): Accuracy 7.67%, latency per test 7,675 s Grok 4.1 Fast (reasoning): Accuracy 3.08%, latency per test 1,533 s Claude Opus 4.5 (non-thinking): Accuracy 23.6%, latency per test 7,682 s Claude Opus 4.5 (thinking): Accuracy 20.3%, latency per test 8,212 s DeepSeek-V3.2 (non-thinking): Accuracy 14.4%, latency per test 2,385 s DeepSeek-V3.2 (thinking): Accuracy 10.7%, latency per test 1,559 s Mistral Large 3: Accuracy 4%, latency per test 6,789 s Devstral 2: Accuracy 0.67%, latency per test 8,045 s Devstral Small 2: Accuracy 2.5%, latency per test 7,467 s GPT-5.2: Accuracy 54.8%, latency per test 6,740 s Gemini 3 Flash: Accuracy 39.1%, latency per test 1,847 s GLM-4.7: Accuracy 7.58%, latency per test 5,317 s MiniMax-M2.1: Accuracy 2.33%, latency per test 5,210 s Kimi K2.5 (thinking): Accuracy 17.7%, latency per test 3,828 s GPT-5.3 Codex: Accuracy 43.8%, latency per test 3,574 s GLM-5 (thinking): Accuracy 22%, latency per test 6,807 s MiniMax-M2.5: Accuracy 6.67%, latency per test 23K s GPT-5.4 (xhigh): Accuracy 67.8%, latency per test 2,661 s GPT-5.4 Mini: Accuracy 6.42%, latency per test 1,897 s GPT-5.4 Nano: Accuracy 15.3%, latency per test 1,171 s MiniMax-M2.7: Accuracy 4.92%, latency per test 3,243 s Claude Opus 4.7: Accuracy 47.1%, latency per test 3,082 s DeepSeek-V4-Pro: Accuracy 35.8%, latency per test 5,374 s Qwen3.7-Max: Accuracy 46.8%, latency per test 4,380 s Claude Fable 5: Accuracy 72.3%, latency per test 3,319 s GPT-5.6 Luna: Accuracy 72.9%, latency per test 1,671 s GPT-5.6 Terra: Accuracy 65.3%, latency per test 1,331 s Qwen 3.8 Max: Accuracy 73%, latency per test 3,946 s Muse Spark 1.2: Accuracy 49.5%, latency per test 4,772 s DeepSeek-V3: Accuracy 1.67%, latency per test 171 s Gemini 2.5 Flash (non-thinking): Accuracy 2.61%, latency per test 350 s GPT-5 mini: Accuracy 6.75%, latency per test 390 s Gemini 3 Pro: Accuracy 38.8%, latency per test 900 s Grok 4.20 (reasoning): Accuracy 30.2%, latency per test 897 s Grok 4.3: Accuracy 15.3%, latency per test 629 s GPT-5.6 Sol: Accuracy 86.7%, latency per test 1,311 s Gemini 3.5 Flash-Lite: Accuracy 26.2%, latency per test 882 s Claude Opus 5: Accuracy 91.7%, latency per test 1,693 s Claude Opus 5GPT-5.6 SolGemini 3 ProGrok 4.20 (reasoning)Gemini 3.5 Flash-LiteGrok 4.3GPT-5 miniGemini 2.5 Flash (non-thinking)DeepSeek-V3
ModelSettingAccuracyLatency per test
Claude Opus 5 91.7% 1,693 s
GPT-5.6 Sol 86.7% 1,311 s
Qwen 3.8 Max 73% 3,946 s
GPT-5.6 Luna 72.9% 1,671 s
Claude Fable 5 72.3% 3,319 s
GPT-5.4 xhigh 67.8% 2,661 s
GPT-5.6 Terra 65.3% 1,331 s
GPT-5.2 54.8% 6,740 s
Muse Spark 1.2 49.5% 4,772 s
Claude Opus 4.7 47.1% 3,082 s
Qwen3.7-Max 46.8% 4,380 s
GPT-5.3 Codex 43.8% 3,574 s
Gemini 3 Flash 39.1% 1,847 s
Gemini 3 Pro 38.8% 900.44 s
DeepSeek-V4-Pro 35.8% 5,374 s
Grok 4.20 reasoning 30.2% 896.93 s
Gemini 3.5 Flash-Lite 26.2% 882.37 s
Grok 4 26.2% 4,266 s
Claude Opus 4.5 non-thinking 23.6% 7,682 s
GLM-5 thinking 22% 6,807 s
GPT-5.1 21.5% 1,750 s
GPT-5.1-Codex-Max 21.4% 956.69 s
Claude Opus 4.5 thinking 20.3% 8,212 s
GPT-5 20% 2,608 s
Claude Sonnet 4.5 thinking 18.3% 3,315 s
Kimi K2.5 thinking 17.7% 3,828 s
Gemini 2.5 Pro Jun 2025 17.1% 2,965 s
Qwen3-Max 15.7% 2,045 s
Grok 4.3 15.3% 628.82 s
GPT-5.4 Nano 15.3% 1,171 s
DeepSeek-V3.2 non-thinking 14.4% 2,385 s
Qwen3-Max 13.8% 2,953 s
Claude Opus 4.1 non-thinking 12.5% 6,125 s
Grok 4 Fast reasoning 11.5% 1,409 s
DeepSeek-V3.2 thinking 10.7% 1,559 s
GPT-5-Codex 9.8% 1,543 s
Qwen3-Max 7.8% 3,372 s
Grok 4.1 Fast non-reasoning 7.7% 7,675 s
GLM-4.7 7.6% 5,317 s
GPT-5 mini 6.8% 390.38 s
MiniMax-M2.5 6.7% 22,888 s
Claude Sonnet 4 non-thinking 6.5% 3,253 s
GPT-5.4 Mini 6.4% 1,897 s
Claude Haiku 4.5 thinking 6.2% 7,415 s
MiniMax-M2.7 4.9% 3,243 s
o4-mini 4.8% 480.19 s
Claude Sonnet 4 thinking 4.6% 2,216 s
grok-code-fast-1 4.3% 1,084 s
GLM-4.6 4.3% 7,106 s
Mistral Large 3 4% 6,789 s
Grok 4 Fast non-reasoning 3.8% 577.63 s
GPT-5.1-Codex 3.7% 2,227 s
Grok 4.1 Fast reasoning 3.1% 1,533 s
GLM-4.5 2.9% 2,336 s
Gemini 2.5 Flash Jun 2025 non-thinking 2.6% 350.19 s
Devstral Small 2 2.5% 7,467 s
MiniMax-M2.1 2.3% 5,210 s
DeepSeek-V3 Mar 2025 1.7% 171.48 s
Kimi K2 Jul 2025 1.3% 1,551 s
Magistral Medium 1.2 0.7% 2,212 s
Devstral 2 0.7% 8,045 s
Qwen3-235B-A22B 0% 352.56 s

The best score, 91.7%, costs $5.54 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 Anthropic Qwen Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5 GPT-5.6 Sol Qwen 3.8 Max GPT-5.6 Luna GPT-5.4 Nano GPT-5 mini 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5 Qwen 3.8 Max

Source: Official IOI (Vals) leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5 91.7% $5.54
GPT-5.6 Sol 86.7% $2.98
Qwen 3.8 Max 73% $2.39
GPT-5.6 Luna 72.9% $0.26
Claude Fable 5 72.3% $15
GPT-5.4 xhigh 67.8% $3.47 127,091
GPT-5.6 Terra 65.3% $0.97
GPT-5.2 54.8% $3.87 142,786
Muse Spark 1.2 49.5% $6.10
Claude Opus 4.7 47.1% $10 227,221
Qwen3.7-Max 46.8% $4.86 287,807
GPT-5.3 Codex 43.8% $2.02 96,713
Gemini 3 Flash 39.1% $0.38 70,681
Gemini 3 Pro 38.8% $0.33 10,237
DeepSeek-V4-Pro 35.8% $0.57 189,618
Grok 4.20 reasoning 30.2% $0.71 75,424
Grok 4 26.2% $2.69 86,145
Gemini 3.5 Flash-Lite 26.2% $0.64
Claude Opus 4.5 non-thinking 23.6% $3.05
GLM-5 thinking 22% $3.27 112,528
GPT-5.1 21.5% $0.37 13,857
GPT-5.1-Codex-Max 21.4% $0.42 10,177
Claude Opus 4.5 thinking 20.3% $3.92 156,866
GPT-5 20% $1.63 128,447
Claude Sonnet 4.5 thinking 18.3% $3.07 44,427
Kimi K2.5 thinking 17.7% $0.68 134,793
Gemini 2.5 Pro Jun 2025 17.1% $0.62 26,594
Qwen3-Max 15.7% $0.82 33,377
Grok 4.3 15.3% $0.35 64,107
GPT-5.4 Nano 15.3% $0.04 86,687
DeepSeek-V3.2 non-thinking 14.4% $0.05 29,512
Qwen3-Max 13.8% $1.90 93,580
Claude Opus 4.1 non-thinking 12.5% $18 34,971
Grok 4 Fast reasoning 11.5% $0.97 1,413,897
DeepSeek-V3.2 thinking 10.7% $0.06 18,056
GPT-5-Codex 9.75% $2.67 87,592
Qwen3-Max 7.75% $1.40 37,429
Grok 4.1 Fast non-reasoning 7.67% $0.07 30,720
GLM-4.7 7.58% $0.44 48,323
GPT-5 mini 6.75% $0.04 10,849
MiniMax-M2.5 6.67% $1.50 86,409
Claude Sonnet 4 non-thinking 6.50% $2.39 23,289
GPT-5.4 Mini 6.42% $0.35 132,677
Claude Haiku 4.5 thinking 6.17% $2.13 69,392
MiniMax-M2.7 4.92% $1.16 109,783
o4-mini 4.83% $0.16 24,353
Claude Sonnet 4 thinking 4.58% $1.33 16,334
grok-code-fast-1 4.33% $1.62 962,152
GLM-4.6 4.33% $1.52 65,549
Mistral Large 3 4.00% $3.07 139,222
Grok 4 Fast non-reasoning 3.83% $0.13 184,454
GPT-5.1-Codex 3.67% $0.18 12,151
Grok 4.1 Fast reasoning 3.08% $0.03 10,191
GLM-4.5 2.92% $0.94 75,865
Gemini 2.5 Flash Jun 2025 non-thinking 2.61% $0.04 7,769
Devstral Small 2 2.50% $0.51 88,182
MiniMax-M2.1 2.33% $0.32 82,905
DeepSeek-V3 Mar 2025 1.67% $0.61 7,904
Kimi K2 Jul 2025 1.25% $2.37 36,790
Magistral Medium 1.2 0.67% $6.34 82,622
Devstral 2 0.67% $1.75 62,865
Qwen3-235B-A22B 0% $0.11 30,567

7 models have held number one on the official leaderboard since its release in August 2025

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

xAI Google DeepMind OpenAI Anthropic
Own leaderboard Passed 85% in Jul 2026, by GPT-5.6 Sol
Claude Opus 5 91.7%

Source: Official IOI (Vals) leaderboard

GPT-5.6 Sol on the official leaderboard passed 85% in July 2026

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 Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Claude Opus 5 GPT-5.6 Sol Claude Fable 5 GPT-5.4 GPT-5.2 Gemini 3 Pro Grok 4 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Claude Opus 5 Claude Fable 5

Source: Official IOI (Vals) leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

xAI Google DeepMind OpenAI Anthropic
IOI v2 (Vals) Passed 85% in Oct 2026, by GPT-6 Astra
GPT-6 Astra 100%
IOI (Vals) Passed 85% in Jul 2026, by GPT-5.6 Sol
Claude Opus 5 91.7%
IOI v2 (Vals) Passed 85% in Oct 2026, by GPT-6 Astra
GPT-6 Astra 100%

Source: Official leaderboards

Replaced by IOI v2 (Vals)

More in Coding

All 282 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471