Benchmark Almanac

MultiChallenge

GPT-6.1 Sol 82% leads since September 2026
GPT-6.1 Sol 82% leads since September 2026
In use

Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics.

On February 3rd, 2026 Scale switched the judge to Gemini 2.5 Pro and rewrote the rubric questions of 54 tasks. Scores published before that date may not match the current board.

Made by
Scale AI
Released
January 2025
Kind
Multi-turn conversation
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
Public, plus a private test set
Strongest model tested
GPT-6.1 Sol #5 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026

In the authors’ words

We present MultiChallenge, a pioneering benchmark evaluating large language models (LLMs) on conducting multi-turn conversations with human users, a crucial yet underexamined capability for their applications. MultiChallenge identifies four categories of challenges in multi-turn conversations that are not only common and realistic among current human-LLM interactions, but are also challenging to all current frontier LLMs. All 4 challenges require accurate instruction-following, context allocation, and in-context reasoning at the same time. We also develop LLM as judge with instance-level rubrics to facilitate an automatic evaluation method with fair agreement with experienced human raters. Despite achieving near-perfect scores on existing multi-turn evaluation benchmarks, all frontier models have less than 50% accuracy on MultiChallenge, with the top-performing Claude 3.5 Sonnet (June 2024) achieving just a 41.4% average accuracy.

Abstract of MultiChallenge: A Realistic Multi-Turn Conversation Evaluation Benchmark Challenging to Frontier LLMs (arXiv:2501.17399, CC0)
44
Entries on its own leaderboard, 42 models we track
36
Launch posts list it, from 13 labs
21
System cards report it, from 11 labs

GPT-6.1 Sol leads with 82%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. 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.

GPT-6.1 Sol leads with 82%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

44 entries. Scroll sideways for the rest.

  1. 82.0 GPT-6.1 Sol
  2. 78.7 Gemini 4 Argon
  3. 75.5 Muse Spark
  4. 75.3 Muse Spark 1.1
  5. 74.9 DeepSeek V4.1 Flash 1017
  6. 74.2 Muse Spark Open
  7. 74.2 GPT-6.1 Terra
  8. 71.4 Gemini 3.1 Pro
  9. 69.2 GPT-5.4 Pro
  10. 67.7 Claude Haiku 5.5
  11. 65.7 Gemini 3 Pro
  12. 63.4 GPT-5.1
  13. 63.2 GPT-5
  14. 62.4 o3-pro
  15. 61.4 Kimi K2.5
  16. 60.6 Gemini 3.1 Flash-Lite
  17. 59.0 GPT-5 mini
  18. 59.0 Claude Opus 4.5
  19. 58.6 Claude Opus 4
  20. 57.2 Claude Opus 4.1
  21. 57.1 Claude Sonnet 4
  22. 56.6 o3
  23. 56.0 Nemotron 3.5 Bolt E4B
  24. 56.0 Claude Opus 4.6
  25. 55.4 Kimi K2 Thinking
  26. 55.3 Claude Sonnet 4.5
  27. 53.6 Gemini 2.5 Pro
  28. 51.6 Claude 3.7 Sonnet
  29. 51.2 GPT-5.1 Instant
  30. 50.5 Claude Haiku 4.5
  31. 46.1 DeepSeek-V3.1
  32. 45.3 gpt-oss-120b
  33. 44.9 o4-mini
  34. 43.2 Gemma 4 E4B
  35. 41.2 Qwen3-235B-A22B
  36. 39.4 GPT-4.1
  37. 38.7 Nemotron 3 Nano 4B
  38. 36.4 Gemini 2.0 Flash
  39. 36.0 Gemma 4 E2B
  40. 27.8 Ministral 3 8B
  41. 25.7 Ministral 3 3B
  42. 25.1 Qwen3.5-4B
  43. 23.7 Llama 3.1-8B
  44. 22.8 Gemma 3 4B
  1. 1 GPT-6.1 Sol 82%
  2. 2 Gemini 4 Argon 78.7%
  3. 3 Muse Spark 75.5%
  4. 4 Muse Spark 1.1 75.3%
  5. 5 DeepSeek V4.1 Flash 1017 74.9%
  6. 6 Muse Spark Open 74.2%
  7. 7 GPT-6.1 Terra 74.2%
  8. 8 Gemini 3.1 Pro 71.4%
  9. 9 GPT-5.4 Pro 69.2%
  10. 10 Claude Haiku 5.5 67.7%
  11. 11 Gemini 3 Pro 65.7%
  12. 12 GPT-5.1 thinking 63.4%
  13. 13 GPT-5 thinking 63.2%
  14. 14 o3-pro high 62.4%
  15. 15 Kimi K2.5 61.4%
  16. 16 Gemini 3.1 Flash-Lite 60.6%
  17. 17 GPT-5 mini thinking 59%
  18. 18 Claude Opus 4.5 thinking 59%
  19. 19 Claude Opus 4 thinking 58.6%
  20. 20 Claude Opus 4.1 thinking 57.2%
  21. 21 Claude Sonnet 4 thinking 57.1%
  22. 22 o3 high 56.6%
  23. 23 Nemotron 3.5 Bolt E4B thinking 56%
  24. 24 Claude Opus 4.6 non-thinking 56%
  25. 25 Kimi K2 Thinking 55.4%
  26. 26 Claude Sonnet 4.5 thinking 55.3%
  27. 27 Gemini 2.5 Pro Jun 2025 53.6%
  28. 28 Claude 3.7 Sonnet thinking 51.6%
  29. 29 GPT-5.1 Instant 51.2%
  30. 30 Claude Haiku 4.5 thinking 50.5%
  31. 31 DeepSeek-V3.1 46.1%
  32. 32 gpt-oss-120b 45.3%
  33. 33 o4-mini high 44.9%
  34. 34 Gemma 4 E4B thinking 43.2%
  35. 35 Qwen3-235B-A22B 41.2%
  36. 36 GPT-4.1 39.4%
  37. 37 Nemotron 3 Nano 4B thinking 38.7%
  38. 38 Gemini 2.0 Flash Feb 2025 36.4%
  39. 39 Gemma 4 E2B thinking 36%
  40. 40 Ministral 3 8B thinking 27.8%
  41. 41 Ministral 3 3B thinking 25.7%
  42. 42 Qwen3.5-4B thinking 25.1%
  43. 43 Llama 3.1-8B 23.7%
  44. 44 Gemma 3 4B 22.8%

Source: Official MultiChallenge leaderboard

77 models. Scroll sideways for the rest.

  1. 82.0 GPT-6.1 Sol
  2. 78.7 Gemini 4 Argon
  3. 77.7 Nova 2.0 Pro Preview
  4. 76.6 Nova 2 Lite
  5. 75.5 Muse Spark
  6. 75.5 Nova 2 Omni Preview
  7. 75.3 Muse Spark 1.1
  8. 74.9 DeepSeek V4.1 Flash 1017
  9. 74.2 Muse Spark Open
  10. 74.2 GPT-6.1 Terra
  11. 71.4 Gemini 3.1 Pro
  12. 69.9 Seed2.0 Lite (0428)
  13. 69.8 Ling-3.1-flash
  14. 69.6 GPT-5
  15. 69.2 GPT-5.4 Pro
  16. 68.3 Seed2.0 Pro
  17. 67.7 Claude Haiku 5.5
  18. 67.6 Qwen3.5 397B-A17B
  19. 66.7 Seed1.8
  20. 66.0 ERNIE 5.0
  21. 65.7 Gemini 3 Pro
  22. 64.8 GLM-5
  23. 63.8 Nemotron 3 Ultra
  24. 63.4 GPT-5.1
  25. 63.3 Qwen3-Max
  26. 62.6 STEP3-VL-10B
  27. 62.6 Tencent HY 2.0 Think
  28. 62.4 o3-pro
  29. 62.3 GPT-5 mini
  30. 61.5 Qwen3.5-122B-A10B
  31. 61.4 Kimi K2.5
  32. 61.0 Solar Open 2
  33. 60.8 Qwen3.5-27B
  34. 60.6 Gemini 3.1 Flash-Lite
  35. 60.4 Mercury 3
  36. 60.0 Qwen3.5-35B-A3B
  37. 59.0 Claude Opus 4.5
  38. 58.6 Claude Opus 4
  39. 57.2 Claude Opus 4.1
  40. 57.1 Claude Sonnet 4
  41. 56.6 o3
  42. 56.0 Nemotron 3.5 Bolt E4B
  43. 56.0 Claude Opus 4.6
  44. 55.7 Step 3.5 Flash
  45. 55.4 Kimi K2 Thinking
  46. 55.3 Claude Sonnet 4.5
  47. 55.2 Nemotron 3 Super
  48. 54.9 GPT-5 nano
  49. 54.5 Qwen3.5-9B
  50. 54.1 Kimi K2
  51. 53.6 Gemini 2.5 Pro
  52. 53.0 MAI-Thinking-1
  53. 52.8 GLM-4.5
  54. 51.6 Claude 3.7 Sonnet
  55. 51.2 GPT-5.1 Instant
  56. 50.9 Ring-1T
  57. 50.5 Claude Haiku 4.5
  58. 49.0 Qwen3.5-4B
  59. 46.1 DeepSeek-V3.1
  60. 45.3 gpt-oss-120b
  61. 45.3 Nemotron-Cascade 2 30B-A3B
  62. 44.9 o4-mini
  63. 44.7 MiniMax-M1-80k
  64. 43.2 Gemma 4 E4B
  65. 42.5 GLM-4.5-Air
  66. 41.2 Qwen3-235B-A22B
  67. 39.4 GPT-4.1
  68. 38.7 Nemotron 3 Nano 4B
  69. 38.5 Nemotron 3 Nano 30B-A3B
  70. 36.4 Gemini 2.0 Flash
  71. 36.0 Gemma 4 E2B
  72. 35.8 GPT-4.1 mini
  73. 33.7 Qwen3.5-2B
  74. 23.7 Llama 3.1-8B
  75. 22.8 Gemma 3 4B
  76. 18.9 Qwen3.5-0.8B
  77. 15.0 GPT-4.1 nano
  1. 1 GPT-6.1 Sol 82%
  2. 2 Gemini 4 Argon 78.7%
  3. 3 Nova 2.0 Pro Preview 77.7%
  4. 4 Nova 2 Lite 76.6%
  5. 5 Muse Spark 75.5%
  6. 6 Nova 2 Omni Preview 75.5%
  7. 7 Muse Spark 1.1 75.3%
  8. 8 DeepSeek V4.1 Flash 1017 74.9%
  9. 9 Muse Spark Open 74.2%
  10. 10 GPT-6.1 Terra 74.2%
  11. 11 Gemini 3.1 Pro 71.4%
  12. 12 Seed2.0 Lite (0428) 69.9%
  13. 13 Ling-3.1-flash 69.8%
  14. 14 GPT-5 69.6%
  15. 15 GPT-5.4 Pro 69.2%
  16. 16 Seed2.0 Pro 68.3%
  17. 17 Claude Haiku 5.5 67.7%
  18. 18 Qwen3.5 397B-A17B 67.6%
  19. 19 Seed1.8 66.7%
  20. 20 ERNIE 5.0 66%
  21. 21 Gemini 3 Pro 65.7%
  22. 22 GLM-5 64.8%
  23. 23 Nemotron 3 Ultra 63.8%
  24. 24 GPT-5.1 thinking 63.4%
  25. 25 Qwen3-Max 63.3%
  26. 26 STEP3-VL-10B 62.6%
  27. 27 Tencent HY 2.0 Think 62.6%
  28. 28 o3-pro high 62.4%
  29. 29 GPT-5 mini 62.3%
  30. 30 Qwen3.5-122B-A10B 61.5%
  31. 31 Kimi K2.5 61.4%
  32. 32 Solar Open 2 61%
  33. 33 Qwen3.5-27B 60.8%
  34. 34 Gemini 3.1 Flash-Lite 60.6%
  35. 35 Mercury 3 60.4%
  36. 36 Qwen3.5-35B-A3B 60%
  37. 37 Claude Opus 4.5 thinking 59%
  38. 38 Claude Opus 4 thinking 58.6%
  39. 39 Claude Opus 4.1 thinking 57.2%
  40. 40 Claude Sonnet 4 thinking 57.1%
  41. 41 o3 high 56.6%
  42. 42 Nemotron 3.5 Bolt E4B thinking 56%
  43. 43 Claude Opus 4.6 non-thinking 56%
  44. 44 Step 3.5 Flash 55.7%
  45. 45 Kimi K2 Thinking 55.4%
  46. 46 Claude Sonnet 4.5 thinking 55.3%
  47. 47 Nemotron 3 Super 55.2%
  48. 48 GPT-5 nano 54.9%
  49. 49 Qwen3.5-9B 54.5%
  50. 50 Kimi K2 Jul 2025 54.1%
  51. 51 Gemini 2.5 Pro Jun 2025 53.6%
  52. 52 MAI-Thinking-1 53%
  53. 53 GLM-4.5 52.8%
  54. 54 Claude 3.7 Sonnet thinking 51.6%
  55. 55 GPT-5.1 Instant 51.2%
  56. 56 Ring-1T 50.9%
  57. 57 Claude Haiku 4.5 thinking 50.5%
  58. 58 Qwen3.5-4B 49%
  59. 59 DeepSeek-V3.1 46.1%
  60. 60 gpt-oss-120b 45.3%
  61. 61 Nemotron-Cascade 2 30B-A3B 45.3%
  62. 62 o4-mini high 44.9%
  63. 63 MiniMax-M1-80k 44.7%
  64. 64 Gemma 4 E4B thinking 43.2%
  65. 65 GLM-4.5-Air 42.5%
  66. 66 Qwen3-235B-A22B 41.2%
  67. 67 GPT-4.1 39.4%
  68. 68 Nemotron 3 Nano 4B thinking 38.7%
  69. 69 Nemotron 3 Nano 30B-A3B 38.5%
  70. 70 Gemini 2.0 Flash Feb 2025 36.4%
  71. 71 Gemma 4 E2B thinking 36%
  72. 72 GPT-4.1 mini 35.8%
  73. 73 Qwen3.5-2B 33.7%
  74. 74 Llama 3.1-8B 23.7%
  75. 75 Gemma 3 4B 22.8%
  76. 76 Qwen3.5-0.8B 18.9%
  77. 77 GPT-4.1 nano 15%

Sources: Official MultiChallenge leaderboard and labs’ model cards

#EntryScoreSettingDateSource
1 GPT-6.1 Sol 82% Sep 2026 Official leaderboard
2 Gemini 4 Argon 78.7% Sep 2026 Official leaderboard
3 Muse Spark 75.5% Apr 2026 Official leaderboard
4 Muse Spark 1.1 75.3% Jul 2026 Official leaderboard
5 DeepSeek V4.1 Flash 1017 74.9% Oct 2026 Official leaderboard
6 Muse Spark Open 74.2% Oct 2026 Official leaderboard
7 GPT-6.1 Terra 74.2% Oct 2026 Official leaderboard
8 Gemini 3.1 Pro 71.4% Feb 2026 Official leaderboard
9 GPT-5.4 Pro 69.2% Mar 2026 Official leaderboard
10 Claude Haiku 5.5 67.7% Oct 2026 Official leaderboard
11 Gemini 3 Pro 65.7% Nov 2025 Official leaderboard
12 GPT-5.1 63.4% thinking Nov 2025 Official leaderboard
13 GPT-5 63.2% thinking Aug 2025 Official leaderboard
14 o3-pro 62.4% high Jun 2025 Official leaderboard
15 Kimi K2.5 61.4% Jan 2026 Official leaderboard
16 Gemini 3.1 Flash-Lite 60.6% Mar 2026 Official leaderboard
17 GPT-5 mini 59% thinking Aug 2025 Official leaderboard
18 Claude Opus 4.5 59% thinking Nov 2025 Official leaderboard
19 Claude Opus 4 58.6% thinking May 2025 Official leaderboard
20 Claude Opus 4.1 57.2% thinking Aug 2025 Official leaderboard
21 Claude Sonnet 4 57.1% thinking May 2025 Official leaderboard
22 o3 56.6% high Apr 2025 Official leaderboard
23 Nemotron 3.5 Bolt E4B 56% thinking Oct 2026 Official leaderboard
24 Claude Opus 4.6 56% non-thinking Feb 2026 Official leaderboard
25 Kimi K2 Thinking 55.4% Nov 2025 Official leaderboard
26 Claude Sonnet 4.5 55.3% thinking Sep 2025 Official leaderboard
27 Gemini 2.5 Pro Jun 2025 53.6% Jun 2025 Official leaderboard
28 Claude 3.7 Sonnet 51.6% thinking Feb 2025 Official leaderboard
29 GPT-5.1 Instant 51.2% Nov 2025 Official leaderboard
30 Claude Haiku 4.5 50.5% thinking Oct 2025 Official leaderboard
31 DeepSeek-V3.1 46.1% Aug 2025 Official leaderboard
32 gpt-oss-120b 45.3% Aug 2025 Official leaderboard
33 o4-mini 44.9% high Apr 2025 Official leaderboard
34 Gemma 4 E4B 43.2% thinking Mar 2026 Official leaderboard
35 Qwen3-235B-A22B 41.2% Apr 2025 Official leaderboard
36 GPT-4.1 39.4% Apr 2025 Official leaderboard
37 Nemotron 3 Nano 4B 38.7% thinking Mar 2026 Official leaderboard
38 Gemini 2.0 Flash Feb 2025 36.4% Feb 2025 Official leaderboard
39 Gemma 4 E2B 36% thinking Mar 2026 Official leaderboard
40 Ministral 3 8B 27.8% thinking Official leaderboard
41 Ministral 3 3B 25.7% thinking Official leaderboard
42 Qwen3.5-4B 25.1% thinking Feb 2026 Official leaderboard
43 Llama 3.1-8B 23.7% Jul 2024 Official leaderboard
44 Gemma 3 4B 22.8% Mar 2025 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.1 Sol 82% Sep 2026 Official leaderboard
2 Gemini 4 Argon 78.7% Sep 2026 Official leaderboard
3 Nova 2.0 Pro Preview 77.7% Dec 2025 Lab model card
4 Nova 2 Lite 76.6% Dec 2025 Lab model card
5 Muse Spark 75.5% Apr 2026 Official leaderboard
6 Nova 2 Omni Preview 75.5% Dec 2025 Lab model card
7 Muse Spark 1.1 75.3% Jul 2026 Official leaderboard
8 DeepSeek V4.1 Flash 1017 74.9% Oct 2026 Official leaderboard
9 Muse Spark Open 74.2% Oct 2026 Official leaderboard
10 GPT-6.1 Terra 74.2% Oct 2026 Official leaderboard
11 Gemini 3.1 Pro 71.4% Feb 2026 Official leaderboard
12 Seed2.0 Lite (0428) 69.9% Apr 2026 Lab model card
13 Ling-3.1-flash 69.8% Oct 2026 Lab model card
14 GPT-5 69.6% Aug 2025 Lab model card
15 GPT-5.4 Pro 69.2% Mar 2026 Official leaderboard
16 Seed2.0 Pro 68.3% Feb 2026 Lab model card
17 Claude Haiku 5.5 67.7% Oct 2026 Official leaderboard
18 Qwen3.5 397B-A17B 67.6% Feb 2026 Lab model card
19 Seed1.8 66.7% Dec 2025 Lab model card
20 ERNIE 5.0 66% Feb 2026 Lab model card
21 Gemini 3 Pro 65.7% Nov 2025 Official leaderboard
22 GLM-5 64.8% Feb 2026 Lab model card
23 Nemotron 3 Ultra 63.8% Jun 2026 Lab model card
24 GPT-5.1 63.4% thinking Nov 2025 Official leaderboard
25 Qwen3-Max 63.3% Jan 2026 Lab model card
26 STEP3-VL-10B 62.6% Jan 2026 Lab model card
27 Tencent HY 2.0 Think 62.6% Dec 2025 Lab model card
28 o3-pro 62.4% high Jun 2025 Official leaderboard
29 GPT-5 mini 62.3% Aug 2025 Lab model card
30 Qwen3.5-122B-A10B 61.5% Feb 2026 Lab model card
31 Kimi K2.5 61.4% Jan 2026 Official leaderboard
32 Solar Open 2 61% Jul 2026 Lab model card
33 Qwen3.5-27B 60.8% Feb 2026 Lab model card
34 Gemini 3.1 Flash-Lite 60.6% Mar 2026 Official leaderboard
35 Mercury 3 60.4% Oct 2026 Lab model card
36 Qwen3.5-35B-A3B 60% Feb 2026 Lab model card
37 Claude Opus 4.5 59% thinking Nov 2025 Official leaderboard
38 Claude Opus 4 58.6% thinking May 2025 Official leaderboard
39 Claude Opus 4.1 57.2% thinking Aug 2025 Official leaderboard
40 Claude Sonnet 4 57.1% thinking May 2025 Official leaderboard
41 o3 56.6% high Apr 2025 Official leaderboard
42 Nemotron 3.5 Bolt E4B 56% thinking Oct 2026 Official leaderboard
43 Claude Opus 4.6 56% non-thinking Feb 2026 Official leaderboard
44 Step 3.5 Flash 55.7% Feb 2026 Lab model card
45 Kimi K2 Thinking 55.4% Nov 2025 Official leaderboard
46 Claude Sonnet 4.5 55.3% thinking Sep 2025 Official leaderboard
47 Nemotron 3 Super 55.2% Mar 2026 Lab model card
48 GPT-5 nano 54.9% Aug 2025 Lab model card
49 Qwen3.5-9B 54.5% Mar 2026 Lab model card
50 Kimi K2 Jul 2025 54.1% Jul 2025 Lab model card
51 Gemini 2.5 Pro Jun 2025 53.6% Jun 2025 Official leaderboard
52 MAI-Thinking-1 53% Jun 2026 Lab model card
53 GLM-4.5 52.8% Aug 2025 Lab model card
54 Claude 3.7 Sonnet 51.6% thinking Feb 2025 Official leaderboard
55 GPT-5.1 Instant 51.2% Nov 2025 Official leaderboard
56 Ring-1T 50.9% Oct 2025 Lab model card
57 Claude Haiku 4.5 50.5% thinking Oct 2025 Official leaderboard
58 Qwen3.5-4B 49% Mar 2026 Lab model card
59 DeepSeek-V3.1 46.1% Aug 2025 Official leaderboard
60 gpt-oss-120b 45.3% Aug 2025 Official leaderboard
61 Nemotron-Cascade 2 30B-A3B 45.3% Mar 2026 Lab model card
62 o4-mini 44.9% high Apr 2025 Official leaderboard
63 MiniMax-M1-80k 44.7% Jun 2025 Lab model card
64 Gemma 4 E4B 43.2% thinking Mar 2026 Official leaderboard
65 GLM-4.5-Air 42.5% Aug 2025 Lab model card
66 Qwen3-235B-A22B 41.2% Apr 2025 Official leaderboard
67 GPT-4.1 39.4% Apr 2025 Official leaderboard
68 Nemotron 3 Nano 4B 38.7% thinking Mar 2026 Official leaderboard
69 Nemotron 3 Nano 30B-A3B 38.5% Dec 2025 Lab model card
70 Gemini 2.0 Flash Feb 2025 36.4% Feb 2025 Official leaderboard
71 Gemma 4 E2B 36% thinking Mar 2026 Official leaderboard
72 GPT-4.1 mini 35.8% Apr 2025 Lab model card
73 Qwen3.5-2B 33.7% Mar 2026 Lab model card
74 Llama 3.1-8B 23.7% Jul 2024 Official leaderboard
75 Gemma 3 4B 22.8% Mar 2025 Official leaderboard
76 Qwen3.5-0.8B 18.9% Mar 2026 Lab model card
77 GPT-4.1 nano 15% Apr 2025 Lab model card

Labs report scores for 42 models in their cards. Switch the scores above to rank them together.

12 models have held number one on the official leaderboard since its release in January 2025 9 models have held number one with claims included since its release in January 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 and what labs claim in their cards.

Meta Google DeepMind Anthropic OpenAI Amazon
Own leaderboard 12 leaders
GPT-6.1 Sol 82%
Lab claims 4 leaders
Nova 2.0 Pro Preview 77.7%

Sources: Official MultiChallenge leaderboard and labs’ model cards

Scores over time

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 2027 Near the ceiling (85%) GPT-6.1 Sol Nova 2.0 Pro Preview GPT-5 o3-pro o3 Claude 3.7 Sonnet Gemini 2.0 Flash 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-6.1 Sol GPT-5

Sources: Official MultiChallenge leaderboard and labs’ model cards

5 labs put it in a launch post in the last six months

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.

OpenAI
MiniMax
Moonshot AI
Amazon
Tencent
NVIDIA
ByteDance Seed
Qwen
Baidu
Microsoft
Upstage
Ant Group
Inception Labs
DateLaunch postScoreAs the post writes it
Oct 23, 2026 Nemotron 3.5 Bolt E4B 55.6 MultiChallenge
Oct 19, 2026 Mercury 3 60.4 MultiChallenge
Oct 13, 2026 Ling-3.1-flash 69.8 MultiChallenge
Jul 22, 2026 Solar Open 2 61 Multi-Challenge
Jun 4, 2026 Nemotron 3 Ultra 63.8 Multi-Challenge
Jun 2, 2026 MAI-Thinking-1 53 Multi-Challenge
Apr 28, 2026 Seed2.0 Lite (0428) 69.9 MultiChallenge
Mar 19, 2026 Nemotron-Cascade 2 30B-A3B 45.3 Scale AI Multi-Challenge
Mar 11, 2026 Nemotron 3 Super 55.2 Scale AI Multi-Challenge
Mar 2, 2026 Qwen3.5-0.8B 18.9 MultiChallenge
Mar 2, 2026 Qwen3.5-2B 33.7 MultiChallenge
Mar 2, 2026 Qwen3.5-4B 49 MultiChallenge
Mar 2, 2026 Qwen3.5-9B 54.5 MultiChallenge
Feb 24, 2026 Qwen3.5-35B-A3B 60 MultiChallenge
Feb 24, 2026 Qwen3.5-27B 60.8 MultiChallenge
Feb 24, 2026 Qwen3.5-122B-A10B 61.5 MultiChallenge
Feb 16, 2026 Qwen3.5 397B-A17B 67.6 MultiChallenge
Feb 14, 2026 Seed2.0 Pro 68.3 MultiChallenge
Feb 6, 2026 ERNIE 5.0 66 MultiChallenge
Jan 23, 2026 Qwen3-Max 63.3 MultiChallenge
Dec 18, 2025 Seed1.8 66.7 MultiChallenge
Dec 15, 2025 Nemotron 3 Nano 30B-A3B 38.5 Scale AI Multi Challenge
Dec 5, 2025 Tencent HY 2.0 Think 62.6 Multi Challenge
Dec 2, 2025 Nova 2 Omni Preview 75.5 MultiChallenge acc
Dec 2, 2025 Nova 2 Lite 76.6 MultiChallenge acc
Dec 2, 2025 Nova 2.0 Pro Preview 77.7 MultiChallenge acc
Aug 7, 2025 GPT-5 nano 54.9 Scale MultiChallenge (o3-mini grader)
Aug 7, 2025 GPT-5 mini 62.3 Scale MultiChallenge (o3-mini grader)
Aug 7, 2025 GPT-5 69.6 Scale MultiChallenge** Multi-turn instruction following
Jul 11, 2025 Kimi K2 54.1 Multi-Challenge, Acc
Jun 16, 2025 MiniMax-M1-80k 44.7 MultiChallenge
Apr 16, 2025 o4-mini 43 Scale MultiChallenge
Apr 16, 2025 o3 56.5 Scale MultiChallenge Multi-turn instruction following
Apr 14, 2025 GPT-4.1 nano 15 MultiChallenge (GPT-4o grader) 15.0 (o3-mini grader 31.1)
Apr 14, 2025 GPT-4.1 mini 35.8 MultiChallenge (GPT-4o grader) 35.8 (o3-mini grader 42.2)
Apr 14, 2025 GPT-4.1 38.3 MultiChallenge

21 system cards report it, from 11 labs

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
Oct 23, 2026 Nemotron 3.5 Bolt E4B technical report 55.6 MultiChallenge, released Q4_K_M imatrix checkpoint (Table 5). BF16 53.1 (Table 3) Capabilities
Jul 22, 2026 Solar Open 2 technical report 61 Multi-Challenge (Table 3) Capabilities
Jun 9, 2026 Nemotron 3 Ultra technical report 63.8 Multi-Challenge (Table 10) Capabilities
Jun 2, 2026 MAI-Thinking-1 technical report 53 Multi-Challenge, Gemini 2.5 Pro judge (Table 12, rounded) Capabilities
Mar 21, 2026 Seed1.8 model card 66.7 MultiChallenge (Table 1) Capabilities
Mar 19, 2026 Nemotron-Cascade 2 30B-A3B technical report 45.3 Scale AI Multi-Challenge (Table 1) Capabilities
Mar 11, 2026 Nemotron 3 Super technical report 55.2 Scale AI Multi-Challenge (Table 5) Capabilities
Feb 17, 2026 GLM-5 technical report 64.8 Multi Challenge, Instruction Following panel (Figure 11) Capabilities
Feb 14, 2026 Seed2.0 Pro model card 68.3 MultiChallenge (Table 3) Capabilities
Feb 11, 2026 Step 3.5 Flash technical report 55.7 MultiChallenge, o3-mini judge (Table 5). PaCoRe 60.8 Capabilities
Feb 4, 2026 ERNIE 5.0 technical report 66 MultiChallenge (Table 2) Capabilities
Jan 14, 2026 STEP3-VL-10B technical report 62.6 MultiChallenge, SeRe (Table 2) Capabilities
Dec 23, 2025 Nemotron 3 Nano 30B-A3B technical report 38.5 Scale AI Multi Challenge (Table 3) Capabilities
Dec 2, 2025 Nova 2 Omni Preview technical report 75.5 MultiChallenge acc (Table 5) Capabilities
Dec 2, 2025 Nova 2 Lite technical report 76.6 MultiChallenge acc (Table 1) Capabilities
Dec 2, 2025 Nova 2.0 Pro Preview technical report 77.7 MultiChallenge acc (Table 1) Capabilities
Oct 21, 2025 Ring-1T technical report 50.9 MultiChallenge (Table 1) Capabilities
Aug 8, 2025 GLM-4.5 technical report 52.8 MultiChallenge (Table 6) Capabilities
Aug 8, 2025 GLM-4.5-Air technical report 42.5 MultiChallenge (Table 6) Capabilities
Jul 21, 2025 Kimi K2 technical report 54.1 Multi-Challenge (Acc.) (Table 3) Capabilities
Jun 16, 2025 MiniMax-M1-80k technical report 44.7 MultiChallenge, GPT-4o judge (Table 2) Capabilities

The lab's number comes out ahead in 3 of 6 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 1 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Qwen3.5-4B 49% MultiChallenge 25.1% Leaderboard, thinking +24 pts
GPT-5 69.6% Scale MultiChallenge** Multi-turn instruction following 63.2% Leaderboard, thinking +6.4 pts
GPT-5 mini 62.3% Scale MultiChallenge (o3-mini grader) 59% Leaderboard, thinking +3.3 pts
o4-mini 43% Scale MultiChallenge 44.9% Leaderboard, high −1.9 pts
GPT-4.1 38.3% MultiChallenge 39.4% Leaderboard −1.1 pts
Nemotron 3.5 Bolt E4B 55.6% MultiChallenge 56% Leaderboard, thinking −0.4 pts
o3 56.5% Scale MultiChallenge Multi-turn instruction following 56.6% Leaderboard, high Same

More in Instructions and chat

All 81 in Instructions and chat
  • IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy. Gemma 4 31B IT 98.9%
  • IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose). Nemotron-Cascade 2 30B-A3B 82.9%
  • Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
  • AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics. MAI-Thinking-1 85%
  • IFStruct In use Models generate JSON or YAML that must follow a requested schema and phrasing constraints, without constrained decoding. A response passes only if every structural constraint holds. Gemma 4 31B IT 95.9%
  • HANDBOOK.md In use Agents do routine company work with mock email, chat, calendar, issue and commerce tools while following a 20 to 124 page handbook. Programmatic criteria check required and forbidden actions, and a trial passes only if all pass. Claude Opus 5.5 52.3%