NLP Lineage

24 April 2026 · Efficiency and infrastructure

DeepSeek-V4

DeepSeek-V4

DeepSeek-AI

Date
24 April 2026
What the date marks
Model release
Kind
Model
Organisation
DeepSeek
Access
MIT open weights
Other dates
model card 2026-04-27
Threads
In the guided tour
Chapter 8. The weights get out
Learning phases
6. Local models and deployment

Too recent for its influence to be judged: no later milestone in the data descends from it yet. How levels are decided

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

Read the primary source (api-docs.deepseek.com)

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

What it changed

Before it

Million-token context was a closed-model feature.

The problem

Open models could not match frontier context length at usable cost.

What it introduced

A very large sparse MoE with a million-token context under MIT, built on the sparse attention from V3.2.

Why it matters

It brought frontier-scale long context into open weights, and kept downward pressure on frontier pricing.

What modern models inherit

Open weights competing on context length and price.

What came next

Later V4 checkpoints through 2026.

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

DeepSeek-V4 has 61 documented ancestors across 12 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 23 turning points among them.

DeepSeek-V4 has 61 documented ancestors across 12 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

DeepSeek-V4 has 68 recorded ancestors, including indirect ones, across 12 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 24 turning points among them.

DeepSeek-V4 has 68 recorded ancestors, including indirect ones, across 12 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2026DeepSeek-V4This milestone.
  2. 2025DeepSeek-V3.2DeepSeek-V4 built directly on thisChapter 10. Reading more than you remember
  3. 2024DeepSeek-V3DeepSeek-V4 was influenced by thisChapter 8. The weights get out
  4. 2024Llama 3.1 405BDeepSeek-V3 was influenced by thisChapter 8. The weights get out
  5. 2024GPT-4oDeepSeek-V3 was influenced by thisChapter 11. Seeing and hearing
  6. 2024DeepSeek-V2 / MLADeepSeek-V3 built directly on this
    DeepSeek-V3.2 was influenced by this    Chapter 9. Making it run anywhere
  7. 2024Llama 3Llama 3.1 405B built directly on thisChapter 8. The weights get out
  8. 2024GRPODeepSeek-V3 built directly on thisChapter 12. Thinking longer
  9. 2024DeepSeekMoEDeepSeek-V3 built directly on thisChapter 8. The weights get out
  10. 2024DeepSeek LLMGRPO built directly on thisChapter 8. The weights get out
  11. 2023GPT-4VGPT-4o built directly on thisChapter 11. Seeing and hearing
  12. 2023Llama 2 Turning pointDeepSeek LLM was influenced by this
    Llama 3 built directly on this    Chapter 8. The weights get out
  13. 2023Position interpolationLlama 3.1 405B was influenced by thisChapter 10. Reading more than you remember
  14. 2023DPOLlama 3.1 405B built directly on thisChapter 6. Teaching models what people want
  15. 2023Grouped-query attention Turning pointDeepSeek-V2 / MLA was influenced by this
    Llama 3.1 405B built directly on this    Chapter 9. Making it run anywhere
  16. 2023GPT-4 Turning pointGPT-4o built directly on thisChapter 7. The assistant goes public
  17. 2023LLaMA Turning pointPosition interpolation built directly on thisChapter 8. The weights get out
  18. 2022GPT-3.5 Turning pointGPT-4 succeeded thisChapter 7. The assistant goes public
  19. 2022Chinchilla Turning pointLlama 3.1 405B was influenced by thisChapter 5. Scale becomes a science
  20. 2022InstructGPT Turning pointGRPO was influenced by thisChapter 6. Teaching models what people want
  21. 2021GopherChinchilla built directly on thisChapter 5. Scale becomes a science
  22. 2021FLANInstructGPT was influenced by thisChapter 6. Teaching models what people want
  23. 2021RoPE Turning pointDeepSeek-V2 / MLA built directly on thisChapter 10. Reading more than you remember
  24. 2021Switch TransformerDeepSeekMoE was influenced by thisChapter 5. Scale becomes a science
  25. 2020RLHF for summarisation Turning pointInstructGPT built directly on this
    DPO built directly on this    Chapter 6. Teaching models what people want
  26. 2020GShard Turning pointDeepSeekMoE built directly on thisChapter 5. Scale becomes a science
  27. 2020GPT-3 Turning pointInstructGPT built directly on thisChapter 5. Scale becomes a science
  28. 2020Scaling laws Turning pointChinchilla built directly on this
    GPT-4 built directly on this
    DeepSeek LLM built directly on this    Chapter 5. Scale becomes a science
  29. 2019Multi-query attentionDeepSeek-V2 / MLA was influenced by thisChapter 9. Making it run anywhere
  30. 2019T5 Turning pointSwitch Transformer built directly on thisChapter 4. Pretrain once, use everywhere
  31. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  32. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  33. 2019Sparse Transformer Turning pointDeepSeek-V3.2 was influenced by thisChapter 4. Pretrain once, use everywhere
  34. 2019GPT-2 Turning pointT5 was influenced by this
    Scaling laws was influenced by this
    GPT-3 built directly on this    Chapter 4. Pretrain once, use everywhere
  35. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  36. 2019Transformer-XLRoPE was influenced by thisChapter 4. Pretrain once, use everywhere
  37. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  38. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  39. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  40. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  41. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  42. 2017Mixed precisionDeepSeek-V3 was influenced by thisChapter 9. Making it run anywhere
  43. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  44. 2017Transformer Turning pointSparse Transformer built directly on thisChapter 3. Attention, then only attention
  45. 2017RLHF originRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  46. 2017Sparse MoE Turning pointDeepSeekMoE was influenced by thisChapter 3. Attention, then only attention
  47. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  48. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  49. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  50. 2015BPE tokenizationLLaMA built directly on thisChapter 2. Words become vectors
  51. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  52. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  53. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  54. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  55. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  56. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  57. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  58. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  59. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  60. 1997LSTM Turning pointSparse MoE built directly on thisChapter 1. Language as probability
  61. 1991Mixture of expertsSparse MoE built directly on thisChapter 1. Language as probability
  62. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2026DeepSeek-V4This milestone.
  2. 2025DeepSeek-V3.2DeepSeek-V4 built directly on thisChapter 10. Reading more than you remember
  3. 2024DeepSeek-V3DeepSeek-V4 was influenced by thisChapter 8. The weights get out
  4. 2024Llama 3.1 405BDeepSeek-V3 was influenced by thisChapter 8. The weights get out
  5. 2024GPT-4oDeepSeek-V3 was influenced by thisChapter 11. Seeing and hearing
  6. 2024DeepSeek-V2 / MLADeepSeek-V3 built directly on this
    DeepSeek-V3.2 was influenced by this    Chapter 9. Making it run anywhere
  7. 2024Llama 3Llama 3.1 405B built directly on thisChapter 8. The weights get out
  8. 2024GRPODeepSeek-V3 built directly on thisChapter 12. Thinking longer
  9. 2024DeepSeekMoEDeepSeek-V3 built directly on thisChapter 8. The weights get out
  10. 2024DeepSeek LLMGRPO built directly on thisChapter 8. The weights get out
  11. 2023GPT-4VGPT-4o built directly on thisChapter 11. Seeing and hearing
  12. 2023Llama 2 Turning pointDeepSeek LLM was influenced by this
    Llama 3 built directly on this    Chapter 8. The weights get out
  13. 2023Position interpolationLlama 3.1 405B was influenced by thisChapter 10. Reading more than you remember
  14. 2023Process supervisionGRPO was influenced by this (indirect influence)Chapter 12. Thinking longer
  15. 2023DPOLlama 3.1 405B built directly on thisChapter 6. Teaching models what people want
  16. 2023Grouped-query attention Turning pointDeepSeek-V2 / MLA was influenced by this
    Llama 3.1 405B built directly on this    Chapter 9. Making it run anywhere
  17. 2023GPT-4 Turning pointGPT-4o built directly on thisChapter 7. The assistant goes public
  18. 2023LLaMA Turning pointPosition interpolation built directly on thisChapter 8. The weights get out
  19. 2022GPT-3.5 Turning pointGPT-4 succeeded thisChapter 7. The assistant goes public
  20. 2022WhisperGPT-4o was influenced by this (indirect influence)Chapter 11. Seeing and hearing
  21. 2022Chinchilla Turning pointLlama 3.1 405B was influenced by thisChapter 5. Scale becomes a science
  22. 2022STaRProcess supervision was influenced by this (indirect influence)Chapter 12. Thinking longer
  23. 2022InstructGPT Turning pointGRPO was influenced by thisChapter 6. Teaching models what people want
  24. 2022Chain of thought Turning pointProcess supervision built directly on thisChapter 12. Thinking longer
  25. 2021GopherChinchilla built directly on thisChapter 5. Scale becomes a science
  26. 2021FLANInstructGPT was influenced by thisChapter 6. Teaching models what people want
  27. 2021RoPE Turning pointDeepSeek-V2 / MLA built directly on thisChapter 10. Reading more than you remember
  28. 2021CLIPGPT-4V was influenced by this (indirect influence)Chapter 11. Seeing and hearing
  29. 2021Switch TransformerDeepSeekMoE was influenced by thisChapter 5. Scale becomes a science
  30. 2020Vision TransformerCLIP built directly on thisChapter 11. Seeing and hearing
  31. 2020RLHF for summarisation Turning pointInstructGPT built directly on this
    DPO built directly on this    Chapter 6. Teaching models what people want
  32. 2020GShard Turning pointDeepSeekMoE built directly on thisChapter 5. Scale becomes a science
  33. 2020GPT-3 Turning pointInstructGPT built directly on thisChapter 5. Scale becomes a science
  34. 2020Scaling laws Turning pointChinchilla built directly on this
    GPT-4 built directly on this
    DeepSeek LLM built directly on this    Chapter 5. Scale becomes a science
  35. 2019Multi-query attentionDeepSeek-V2 / MLA was influenced by thisChapter 9. Making it run anywhere
  36. 2019T5 Turning pointSwitch Transformer built directly on thisChapter 4. Pretrain once, use everywhere
  37. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  38. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  39. 2019Sparse Transformer Turning pointDeepSeek-V3.2 was influenced by thisChapter 4. Pretrain once, use everywhere
  40. 2019GPT-2 Turning pointT5 was influenced by this
    Scaling laws was influenced by this
    GPT-3 built directly on this    Chapter 4. Pretrain once, use everywhere
  41. 2019WebTextGPT-2 was influenced by this
    C4 was influenced by this (indirect influence)    Chapter 4. Pretrain once, use everywhere
  42. 2019Transformer-XLSparse Transformer was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  43. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  44. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  45. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  46. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  47. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  48. 2017Mixed precisionDeepSeek-V3 was influenced by thisChapter 9. Making it run anywhere
  49. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  50. 2017Transformer Turning pointSparse Transformer built directly on thisChapter 3. Attention, then only attention
  51. 2017RLHF originRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  52. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  53. 2017Sparse MoE Turning pointDeepSeekMoE was influenced by thisChapter 3. Attention, then only attention
  54. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  55. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  56. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  57. 2015BPE tokenizationLLaMA built directly on thisChapter 2. Words become vectors
  58. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  59. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  60. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  61. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  62. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  63. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  64. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  65. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  66. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  67. 1997LSTM Turning pointSparse MoE built directly on thisChapter 1. Language as probability
  68. 1991Mixture of expertsSparse MoE built directly on thisChapter 1. Language as probability
  69. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

No documented descendants are recorded for DeepSeek-V4 without weak links.

No documented descendants are recorded for DeepSeek-V4 without weak links.

No recorded descendants are recorded for DeepSeek-V4.

No recorded descendants are recorded for DeepSeek-V4.

Direct links and their sources

Where it came from

Built directly on

Influenced by

What it led to

None recorded in the sources used.