Benchmark Almanac

IMO-AnswerBench

GLM-5.2 91% leads since Jun 2026 In use

400 olympiad problems with short verifiable answers, robustified against memorization. Scored by answer match.

Part of IMO-Bench (EMNLP 2025) with IMO-ProofBench, IMO-GradingBench and IMO-LeanProofBench. On 2026-02-12 a v2 file fixed ambiguous problems and wrong answers in place, so check which file a score used.

In the authors’ words

Finding the right north-star metrics is highly critical for advancing the mathematical reasoning capabilities of foundation models, especially given that existing evaluations are either too easy or only focus on getting correct short answers. To address these issues, we present IMO-Bench, a suite of advanced reasoning benchmarks, vetted by a panel of top specialists and that specifically targets the level of the International Mathematical Olympiad (IMO), the most prestigious venue for young mathematicians. IMO-AnswerBench first tests models on 400 diverse Olympiad problems with verifiable short answers. IMO-Proof Bench is the next-level evaluation for proof-writing capabilities, which includes both basic and advanced IMO level problems as well as detailed grading guidelines to facilitate automatic grading. These benchmarks played a crucial role in our historic achievement of the gold-level performance at IMO 2025 with Gemini Deep Think (Luong and Lockhart, 2025). Our model achieved 80.0% on IMO-AnswerBench and 65.7% on the advanced IMO-Proof Bench, surpassing the best non-Gemini models by large margins of 6.9% and 42.4% respectively. We also showed that autograders built with Gemini reasoning correlate well with human evaluations and construct IMO-GradingBench, with 1000 human gradings on proofs, to enable further progress in automatic evaluation of long-form answers. We hope that IMO-Bench will help the community towards advancing robust mathematical reasoning and release it at https://imobench.github.io/.

Abstract of Towards Robust Mathematical Reasoning (arXiv:2511.01846, CC0)
Made by Google DeepMind Released Nov 2025 Competition problems with answers All tasks public Last in a launch post Oct 2026 Last model added Oct 2026
Level
Competition The strongest students, in olympiad-style contests
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.
google-deepmind/superhuman 807 stars Paper Dataset imobench.github.io
37
Models scored in lab cards
32
Launch posts list it, from 13 labs
20
System cards report it, from 11 labs

GLM-5.2 leads with 91%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

37 models. Scroll sideways for the rest.

  1. 91.0 GLM-5.2
  2. 90.0 Ring-2.5-1T
  3. 90.0 Qwen3.7-Max
  4. 90.0 Tencent Hy3
  5. 89.8 DeepSeek-V4-Pro
  6. 89.3 Seed2.0 Pro
  7. 88.6 Nemotron 3 Ultra
  8. 88.4 DeepSeek-V4-Flash
  9. 86.8 LongCat-Flash-Thinking-2601
  10. 86.1 Ring-2.6-1T
  11. 86.0 Kimi K2.6
  12. 86.0 Qwen3.7-Plus
  13. 85.4 Step 3.5 Flash
  14. 84.5 DeepSeek-V3.2-Speciale
  15. 84.3 Tencent Hy3 preview
  16. 83.9 Qwen3-Max
  17. 83.8 Qwen 3.6 Plus
  18. 83.8 GLM-5.1
  19. 82.5 GLM-5
  20. 82.0 GLM-4.7
  21. 81.8 Kimi K2.5
  22. 81.8 LongCat-2.0
  23. 80.9 Qwen3.5 397B-A17B
  24. 80.8 Qwen3.6 27B
  25. 79.3 Nemotron-Cascade 2 30B-A3B
  26. 78.9 Qwen 3.6 35B-A3B
  27. 78.6 Kimi K2 Thinking
  28. 78.6 K-EXAONE 2.0
  29. 78.3 DeepSeek-V3.2
  30. 77.3 Intern-S1-Pro
  31. 76.3 Seed1.8
  32. 76.3 K-EXAONE
  33. 75.2 Spark-X2.6-4B
  34. 73.4 Tencent HY 2.0 Think
  35. 65.8 Ling-2.6-1T
  36. 65.5 Qwen3.5-Omni-Plus
  37. 62.1 STEP3-VL-10B
  1. 1 GLM-5.2 91%
  2. 2 Ring-2.5-1T 90%
  3. 3 Qwen3.7-Max 90%
  4. 4 Tencent Hy3 90%
  5. 5 DeepSeek-V4-Pro 89.8%
  6. 6 Seed2.0 Pro 89.3%
  7. 7 Nemotron 3 Ultra 88.6%
  8. 8 DeepSeek-V4-Flash 88.4%
  9. 9 LongCat-Flash-Thinking-2601 86.8%
  10. 10 Ring-2.6-1T 86.1%
  11. 11 Kimi K2.6 86%
  12. 12 Qwen3.7-Plus 86%
  13. 13 Step 3.5 Flash 85.4%
  14. 14 DeepSeek-V3.2-Speciale 84.5%
  15. 15 Tencent Hy3 preview 84.3%
  16. 16 Qwen3-Max 83.9%
  17. 17 Qwen 3.6 Plus 83.8%
  18. 18 GLM-5.1 83.8%
  19. 19 GLM-5 82.5%
  20. 20 GLM-4.7 82%
  21. 21 Kimi K2.5 81.8%
  22. 22 LongCat-2.0 81.8%
  23. 23 Qwen3.5 397B-A17B 80.9%
  24. 24 Qwen3.6 27B 80.8%
  25. 25 Nemotron-Cascade 2 30B-A3B 79.3%
  26. 26 Qwen 3.6 35B-A3B 78.9%
  27. 27 Kimi K2 Thinking 78.6%
  28. 28 K-EXAONE 2.0 78.6%
  29. 29 DeepSeek-V3.2 78.3%
  30. 30 Intern-S1-Pro 77.3%
  31. 31 Seed1.8 76.3%
  32. 32 K-EXAONE 76.3%
  33. 33 Spark-X2.6-4B 75.2%
  34. 34 Tencent HY 2.0 Think 73.4%
  35. 35 Ling-2.6-1T 65.8%
  36. 36 Qwen3.5-Omni-Plus 65.5%
  37. 37 STEP3-VL-10B 62.1%
#ModelScoreSettingDateSource
1 GLM-5.2 91% Jun 2026 Model card
2 Ring-2.5-1T 90% Feb 2026 Model card
3 Qwen3.7-Max 90% May 2026 Model card
4 Tencent Hy3 90% Jul 2026 Model card
5 DeepSeek-V4-Pro 89.8% Apr 2026 Model card
6 Seed2.0 Pro 89.3% Feb 2026 Model card
7 Nemotron 3 Ultra 88.6% Jun 2026 Model card
8 DeepSeek-V4-Flash 88.4% Apr 2026 Model card
9 LongCat-Flash-Thinking-2601 86.8% Jan 2026 Model card
10 Ring-2.6-1T 86.1% Jun 2026 Model card
11 Kimi K2.6 86% Apr 2026 Model card
12 Qwen3.7-Plus 86% Jun 2026 Model card
13 Step 3.5 Flash 85.4% Feb 2026 Model card
14 DeepSeek-V3.2-Speciale 84.5% Dec 2025 Model card
15 Tencent Hy3 preview 84.3% Apr 2026 Model card
16 Qwen3-Max 83.9% Jan 2026 Model card
17 Qwen 3.6 Plus 83.8% Apr 2026 Model card
18 GLM-5.1 83.8% Apr 2026 Model card
19 GLM-5 82.5% Feb 2026 Model card
20 GLM-4.7 82% Dec 2025 Model card
21 Kimi K2.5 81.8% Jan 2026 Model card
22 LongCat-2.0 81.8% Jul 2026 Model card
23 Qwen3.5 397B-A17B 80.9% Feb 2026 Model card
24 Qwen3.6 27B 80.8% Apr 2026 Model card
25 Nemotron-Cascade 2 30B-A3B 79.3% Mar 2026 Model card
26 Qwen 3.6 35B-A3B 78.9% Apr 2026 Model card
27 Kimi K2 Thinking 78.6% Nov 2025 Model card
28 K-EXAONE 2.0 78.6% Jul 2026 Model card
29 DeepSeek-V3.2 78.3% Dec 2025 Model card
30 Intern-S1-Pro 77.3% Feb 2026 Model card
31 Seed1.8 76.3% Dec 2025 Model card
32 K-EXAONE 76.3% Dec 2025 Model card
33 Spark-X2.6-4B 75.2% Oct 2026 Model card
34 Tencent HY 2.0 Think 73.4% Dec 2025 Model card
35 Ling-2.6-1T 65.8% Jun 2026 Model card
36 Qwen3.5-Omni-Plus 65.5% Mar 2026 Model card
37 STEP3-VL-10B 62.1% Jan 2026 Model card

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

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

Moonshot AI DeepSeek Meituan Ant Group Z.ai
Lab claims Passed 85% in Jan 2026, by LongCat-Flash-Thinking-2601
GLM-5.2 91%

The best score passed 85% in Jan 2026

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

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GLM-5.2 Ring-2.5-1T DeepSeek-V3.2-Speciale Kimi K2 Thinking 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GLM-5.2 DeepSeek-V3.2-Speciale

7 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.

Moonshot AI
DeepSeek
Tencent
ByteDance Seed
Z.ai
LG AI Research
Meituan
Qwen
StepFun
Shanghai AI Lab
Ant Group
NVIDIA
iFlytek
DateLaunch postScoreAs the post writes it
22 Oct 2026 Spark-X2.6-4B 75.2 IMO-AnswerBench
31 Jul 2026 K-EXAONE 2.0 78.6 IMO Answer (IMO-AnswerBench)
5 Jul 2026 LongCat-2.0 81.8 IMO-AnswerBench
2 Jul 2026 Tencent Hy3 90 IMOAnswerBench
16 Jun 2026 GLM-5.2 91 IMOAnswerBench
4 Jun 2026 Nemotron 3 Ultra 88.6 IMOAnswerBench (no tools); with tools 92.3
1 Jun 2026 Qwen3.7-Plus 86 IMOAnswerBench
20 May 2026 Qwen3.7-Max 90 IMOAnswerBench
24 Apr 2026 DeepSeek-V4-Pro 89.8 IMOAnswerBench (Pass@1)
23 Apr 2026 Tencent Hy3 preview 84.3 IMO Answer Bench
22 Apr 2026 Qwen3.6 27B 80.8 IMOAnswerBench
20 Apr 2026 Kimi K2.6 86 IMO-AnswerBench
15 Apr 2026 Qwen 3.6 35B-A3B 78.9 IMOAnswerBench
7 Apr 2026 GLM-5.1 83.8 IMOAnswerBench
2 Apr 2026 Qwen 3.6 Plus 83.8 IMOAnswerBench
30 Mar 2026 Qwen3.5-Omni-Plus 65.5 IMOAnswerBench
19 Mar 2026 Nemotron-Cascade 2 30B-A3B 79.3 IMO AnswerBench
16 Feb 2026 Qwen3.5 397B-A17B 80.9 IMOAnswerBench
14 Feb 2026 Seed2.0 Pro 89.3 IMOAnswerBench (no tool)
11 Feb 2026 GLM-5 82.5 IMOAnswerBench
10 Feb 2026 Ring-2.5-1T 90 Best of IMOAnswerBench (Mean@8) 86.69 / Heavy Thinking 90.00
2 Feb 2026 Intern-S1-Pro 77.3 IMO-Answer-Bench
2 Feb 2026 Step 3.5 Flash 85.4 IMOAnswerBench
27 Jan 2026 Kimi K2.5 81.8 IMO-AnswerBench
23 Jan 2026 Qwen3-Max 83.9 IMOAnswerBench
14 Jan 2026 LongCat-Flash-Thinking-2601 86.8 Best of IMO-AnswerBench (Avg@4) 78.6 / Heavy Thinking 86.8
31 Dec 2025 K-EXAONE 76.3 IMO-AnswerBench
22 Dec 2025 GLM-4.7 82 IMOAnswerBench
18 Dec 2025 Seed1.8 76.3 IMO-AnswerBench (w/ code tools)
5 Dec 2025 Tencent HY 2.0 Think 73.4 IMO-AnswerBench
1 Dec 2025 DeepSeek-V3.2 78.3 IMOAnswerBench (Pass@1)
6 Nov 2025 Kimi K2 Thinking 78.6 IMO-AnswerBench, no tools

20 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
10 Sep 2026 DeepSeek V4.1 Flash technical report IMO-AnswerBench (Pass@1) vs reasoning effort, chart only (Figure 12) Capabilities
5 Aug 2026 K-EXAONE 2.0 technical report 78.6 IMO-AnswerBench (Table 6, Reasoning mode) Capabilities
13 Jun 2026 Ring-2.6-1T technical report 86.1 IMO-AnswerBench (Avg@8), xhigh (Table 6). High gives 66.44 Capabilities
13 Jun 2026 Ling-2.6-1T technical report 65.8 IMO-AnswerBench, Mean@8, CoT (Table 5) Capabilities
9 Jun 2026 Nemotron 3 Ultra technical report 88.6 IMOAnswerBench (no tools), 92.3 with tools (Table 10) Capabilities
26 Apr 2026 DeepSeek-V4-Pro technical report 89.8 IMOAnswerBench (Pass@1), DS-V4-Pro Max (Table 6) Capabilities
26 Apr 2026 DeepSeek-V4-Flash technical report 88.4 IMOAnswerBench (Pass@1), DeepSeek-V4-Flash Max (Table 7) Capabilities
17 Apr 2026 Qwen3.5-Omni-Plus technical report 65.5 IMOAnswerBench (Table 4) Capabilities
26 Mar 2026 Intern-S1-Pro technical report 77.3 IMO-Answer-Bench, thinking (Table 2) Capabilities
21 Mar 2026 Seed1.8 model card 76.3 IMO-AnswerBench, w/ code tools (Table 1) Capabilities
19 Mar 2026 Nemotron-Cascade 2 30B-A3B technical report 79.3 IMO AnswerBench (Table 1) Capabilities
17 Feb 2026 GLM-5 technical report 82.5 IMO-AnswerBench (Table 7) Capabilities
14 Feb 2026 Seed2.0 Pro model card 89.3 IMOAnswerBench (no tool) (Table 3) Capabilities
11 Feb 2026 Step 3.5 Flash technical report 85.4 IMO-AnswerBench (Table 5). PaCoRe 88.8, with Python 86.7 (Table 12) Capabilities
30 Jan 2026 Kimi K2.5 technical report 81.8 IMO-AnswerBench (Table 4) Capabilities
23 Jan 2026 LongCat-Flash-Thinking-2601 technical report 86.8 IMO-AnswerBench (Avg@4), w/ tools, heavy mode (Table 2). Default mode 78.6 Capabilities
14 Jan 2026 STEP3-VL-10B technical report 62.1 IMO-AnswerBench, SeRe (Tables 2 and 3). PaCoRe 76.66 Capabilities
5 Jan 2026 K-EXAONE technical report 76.3 IMO-AnswerBench (Table 3, Reasoning mode). Non-reasoning 40.0 (Table 4) Capabilities
1 Dec 2025 DeepSeek-V3.2 technical report 78.3 IMOAnswerBench (Pass@1), DeepSeek-V3.2 Thinking (Table 2) Capabilities
1 Dec 2025 DeepSeek-V3.2-Speciale technical report 84.5 IMOAnswerBench (Pass@1), DeepSeek-V3.2-Speciale (Table 3) Capabilities

More in Math

All 105 in Math
  • AIME 2026 In use Models solve the 30 integer-answer problems from the 2026 AIME I and II math competitions. Scored by exact answer match. GPT-5.5 100%
  • HMMT 2026 In use Models solve the 33 final-answer problems from the February 2026 Harvard-MIT Mathematics Tournament. Scored by exact answer match. GPT-5.5 98.5%
  • MathArena (Apex, IMO, USAMO) In use Evaluates models on newly released math competitions after release. Final-answer or human/LLM-judged proof scores. Claude Opus 4.8 81.3%
  • PolyMath In use Models solve 9,000 parallel math problems in 18 languages across four difficulty levels, from school math to frontier math. Final answers are checked. Qwen3.7-Max 86.5%
  • USAMO 2026 In use Models write full proofs for the six problems of the 2026 USA Mathematical Olympiad. Judges grade each proof from 0 to 7 against rubrics. GPT-5.5 98.2%
  • ArXivMath In use Final-answer research-level math problems extracted monthly from recent arXiv papers and filtered to be self-contained. GPT-6.1 Sol 94.2%