NLP Lineage

21 July 2025 · Reasoning

IMO gold

IMO gold-medal performance

OpenAI; Google DeepMind

Date
21 July 2025
What the date marks
Announcement
Kind
Benchmark
Organisation
OpenAI; Google DeepMind
Access
research, not released
Other dates
OpenAI 2025-07-19; DeepMind certified 2025-07-21
Threads
In the guided tour
Chapter 12. Thinking longer
Learning phases
2. How LLMs are trained

A milestone: 2 later milestones descend directly from it, and its descendants reach 2 of the 10 threads. How levels are decided

Figures quoted here are reported by the organisation that built it, not independently verified.

Read the primary source (deepmind.google)

Full summary in progress. Until then, this page shows the milestone's record from the research data.

What it changed

Before it

Models solved competition problems only with formal tools or heavy scaffolding.

The problem

Olympiad problems need original proofs written in natural language under time limits.

What it introduced

Two labs independently scored 35 of 42, solving five of six problems, in natural language. DeepMind's run was certified by IMO coordinators.

Why it matters

A milestone for mathematical reasoning, though later independent evaluation found much weaker performance on unfamiliar problems.

What modern models inherit

Both the achievement and the caution about what benchmark scores mean.

What came next

IMO-ProofBench and other independent evaluations testing generalisation.

Before and after

The milestones it descends from, and the milestones that descend from it. Work developed alongside it is not counted. Weak links are left out unless you include them.

Before: where it came from

IMO gold has 40 documented ancestors across 9 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 14 turning points among them.

IMO gold has 40 documented ancestors across 9 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

IMO gold has 73 recorded ancestors, including indirect ones, across 12 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 20 turning points among them.

IMO gold has 73 recorded ancestors, including indirect ones, across 12 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2025IMO goldThis milestone.
  2. 2025Gemini 2.5IMO gold built directly on thisChapter 12. Thinking longer
  3. 2024Gemini 2.0Gemini 2.5 built directly on thisChapter 14. Where the threads meet now
  4. 2024o1 Turning pointGemini 2.5 was influenced by thisChapter 12. Thinking longer
  5. 2024Test-time computeIMO gold was influenced by thisChapter 12. Thinking longer
  6. 2024Gemini 1.5Gemini 2.5 was influenced by thisChapter 10. Reading more than you remember
  7. 2023Gemini 1.0Gemini 1.5 built directly on thisChapter 11. Seeing and hearing
  8. 2023Process supervisionTest-time compute was influenced by thisChapter 12. Thinking longer
  9. 2022PaLMGemini 1.0 succeeded thisChapter 5. Scale becomes a science
  10. 2022Self-consistencyTest-time compute built directly on thisChapter 12. Thinking longer
  11. 2022Chain of thought Turning pointo1 was influenced by this
    Self-consistency built directly on this
    Process supervision built directly on this    Chapter 12. Thinking longer
  12. 2020Vision TransformerGemini 1.0 was influenced by thisChapter 11. Seeing and hearing
  13. 2020GShard Turning pointGemini 1.5 was influenced by thisChapter 5. Scale becomes a science
  14. 2020GPT-3 Turning pointChain of thought was influenced by thisChapter 5. Scale becomes a science
  15. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  16. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  17. 2019GPT-2 Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  18. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  19. 2018BERT Turning pointVision Transformer was influenced by thisChapter 4. Pretrain once, use everywhere
  20. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  21. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  22. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  23. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  24. 2017Transformer Turning pointVision Transformer built directly on thisChapter 3. Attention, then only attention
  25. 2017Sparse MoE Turning pointGShard built directly on thisChapter 3. Attention, then only attention
  26. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  27. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  28. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  29. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  30. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  31. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  32. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  33. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  34. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  35. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  36. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  37. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  38. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  39. 1997LSTM Turning pointSparse MoE built directly on thisChapter 1. Language as probability
  40. 1991Mixture of expertsSparse MoE built directly on thisChapter 1. Language as probability
  41. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2025IMO goldThis milestone.
  2. 2025Gemini 2.5IMO gold built directly on thisChapter 12. Thinking longer
  3. 2024Gemini 2.0Gemini 2.5 built directly on thisChapter 14. Where the threads meet now
  4. 2024Computer useGemini 2.0 was influenced by this (indirect influence)Chapter 13. Models that act
  5. 2024o1 Turning pointGemini 2.5 was influenced by thisChapter 12. Thinking longer
  6. 2024Test-time computeIMO gold was influenced by thisChapter 12. Thinking longer
  7. 2024Claude 3.5 SonnetComputer use built directly on thisChapter 14. Where the threads meet now
  8. 2024Claude 3Claude 3.5 Sonnet built directly on thisChapter 14. Where the threads meet now
  9. 2024Gemini 1.5Gemini 2.5 was influenced by thisChapter 10. Reading more than you remember
  10. 2023Gemini 1.0Gemini 1.5 built directly on thisChapter 11. Seeing and hearing
  11. 2023GPT-4VComputer use was influenced by thisChapter 11. Seeing and hearing
  12. 2023Lost in the middleGemini 1.5 was influenced by this (indirect influence)Chapter 10. Reading more than you remember
  13. 2023Function callingComputer use was influenced by this (indirect influence)Chapter 13. Models that act
  14. 2023Process supervisionTest-time compute was influenced by thisChapter 12. Thinking longer
  15. 2023Tree of thoughtsTest-time compute was influenced by this (indirect influence)Chapter 12. Thinking longer
  16. 2023100K contextLost in the middle was influenced by this (indirect influence)Chapter 10. Reading more than you remember
  17. 2023GPT-4 Turning pointGPT-4V built directly on thisChapter 7. The assistant goes public
  18. 2023Claude100K context built directly on thisChapter 7. The assistant goes public
  19. 2023ToolformerFunction calling was influenced by this (indirect influence)Chapter 13. Models that act
  20. 2022Constitutional AIClaude built directly on this
    Claude 3 built directly on this    Chapter 6. Teaching models what people want
  21. 2022GPT-3.5 Turning pointGPT-4 succeeded thisChapter 7. The assistant goes public
  22. 2022ReActFunction calling was influenced by this (indirect influence)Chapter 13. Models that act
  23. 2022HH-RLHFClaude was influenced by thisChapter 6. Teaching models what people want
  24. 2022PaLMGemini 1.0 succeeded thisChapter 5. Scale becomes a science
  25. 2022STaRTest-time compute was influenced by this (indirect influence)Chapter 12. Thinking longer
  26. 2022Self-consistencyTest-time compute built directly on thisChapter 12. Thinking longer
  27. 2022InstructGPT Turning pointGPT-4 built directly on thisChapter 6. Teaching models what people want
  28. 2022Chain of thought Turning pointo1 was influenced by this
    Self-consistency built directly on this
    STaR was influenced by this    Chapter 12. Thinking longer
  29. 2021WebGPTReAct was influenced by this
    Toolformer was influenced by this    Chapter 13. Models that act
  30. 2021HHH assistantHH-RLHF built directly on thisChapter 6. Teaching models what people want
  31. 2021FLANInstructGPT was influenced by thisChapter 6. Teaching models what people want
  32. 2021CLIPGPT-4V was influenced by this (indirect influence)Chapter 11. Seeing and hearing
  33. 2020Vision TransformerGemini 1.0 was influenced by thisChapter 11. Seeing and hearing
  34. 2020RLHF for summarisation Turning pointInstructGPT built directly on this
    WebGPT built directly on this
    HH-RLHF built directly on this    Chapter 6. Teaching models what people want
  35. 2020GShard Turning pointGemini 1.5 was influenced by thisChapter 5. Scale becomes a science
  36. 2020GPT-3 Turning pointChain of thought was influenced by thisChapter 5. Scale becomes a science
  37. 2020RAGLost in the middle was influenced by thisChapter 10. Reading more than you remember
  38. 2020Dense retrievalRAG built directly on thisChapter 10. Reading more than you remember
  39. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  40. 2019BARTRAG built directly on thisChapter 4. Pretrain once, use everywhere
  41. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  42. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  43. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  44. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  45. 2019GPT-2 Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  46. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  47. 2019Transformer-XLSparse Transformer was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  48. 2018BERT Turning pointVision Transformer was influenced by thisChapter 4. Pretrain once, use everywhere
  49. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  50. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  51. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  52. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  53. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  54. 2017RLHF origino1 was influenced by this (indirect influence)Chapter 6. Teaching models what people want
  55. 2017Transformer Turning pointVision Transformer built directly on thisChapter 3. Attention, then only attention
  56. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  57. 2017FAISSRAG was influenced by thisChapter 10. Reading more than you remember
  58. 2017Sparse MoE Turning pointGShard built directly on thisChapter 3. Attention, then only attention
  59. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  60. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  61. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  62. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  63. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  64. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  65. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  66. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  67. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  68. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  69. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  70. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  71. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  72. 1997LSTM Turning pointSparse MoE built directly on thisChapter 1. Language as probability
  73. 1991Mixture of expertsSparse MoE built directly on thisChapter 1. Language as probability
  74. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

IMO gold has 2 documented descendants across 1 chapter, reaching forward to Gemini 3 Deep Think in 2026. Showing all of them.

IMO gold has 2 documented descendants across 1 chapter, reaching forward to Gemini 3 Deep Think in 2026. Showing all of them.

IMO gold has 2 recorded descendants, including indirect ones, across 1 chapter, reaching forward to Gemini 3 Deep Think in 2026. Showing all of them.

IMO gold has 2 recorded descendants, including indirect ones, across 1 chapter, reaching forward to Gemini 3 Deep Think in 2026. Showing all of them.

  1. 2025IMO goldThis milestone.
  2. 2025IMO-ProofBenchwas influenced by IMO goldChapter 12. Thinking longer
  3. 2026Gemini 3 Deep Thinkwas influenced by IMO goldChapter 12. Thinking longer
  1. 2025IMO goldThis milestone.
  2. 2025IMO-ProofBenchwas influenced by IMO goldChapter 12. Thinking longer
  3. 2026Gemini 3 Deep Thinkwas influenced by IMO goldChapter 12. Thinking longer

Direct links and their sources

Where it came from

Built directly on

Influenced by

What it led to

Influenced