DeepSeek V4

References

Bibliography for the DeepSeek V4 report. All entries date-stamped at access time.

Primary sources

• DeepSeek API Docs — DeepSeek V4 Preview Release announcement. https://api-docs.deepseek.com/news/news260424. Accessed 2026-04-26.
• DeepSeek API Docs — Models & Pricing. https://api-docs.deepseek.com/quick_start/pricing. Accessed 2026-04-26.
• Hugging Face — deepseek-ai/DeepSeek-V4-Pro model card. https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro. Accessed 2026-04-26.
• Hugging Face — DeepSeek_V4.pdf tech report (mirrored on the V4-Pro model card). https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf. Accessed 2026-04-26 (existence confirmed; full ingestion pending).
• DeepSeek-AI — DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models. arXiv:2512.02556, December 2025. https://arxiv.org/abs/2512.02556. Accessed 2026-04-26. (Introduces DeepSeek Sparse Attention.)
• DeepSeek-AI — DeepSeek-V3 Technical Report. arXiv:2412.19437, December 2024. https://arxiv.org/abs/2412.19437. Accessed 2026-04-27.
• Hugging Face — deepseek-ai/DeepSeek-V3 model card and config.json. https://huggingface.co/deepseek-ai/DeepSeek-V3/raw/main/config.json. Accessed 2026-04-27. (Used to verify V3 column of the diff table.)
• Z. Xie, Y. Wei, H. Cao, … W. Liang — mHC: Manifold-Constrained Hyper-Connections. arXiv:2512.24880, 2025-12-31 (v1) / 2026-01-05 (v2). https://arxiv.org/abs/2512.24880. Accessed 2026-04-26.
• D. Zhu, H. Huang, Z. Huang, Y. Zeng, Y. Mao, B. Wu, Q. Min, X. Zhou — Hyper-Connections. arXiv:2409.19606, September 2024 (v1) / March 2025 (v3). https://arxiv.org/abs/2409.19606. Accessed 2026-04-27. (The upstream paper that mHC extends — introduces the residual-stream widening concept.)

Press

• CNBC — China’s DeepSeek releases preview of long-awaited V4 model as AI race intensifies. 2026-04-24. https://www.cnbc.com/2026/04/24/deepseek-v4-llm-preview-open-source-ai-competition-china.html
• Bloomberg — DeepSeek Unveils Newest Flagship AI Model a Year after Upending Silicon Valley. 2026-04-24. https://www.bloomberg.com/news/articles/2026-04-24/deepseek-unveils-newest-flagship-a-year-after-ai-breakthrough
• Al Jazeera — China’s DeepSeek unveils latest models a year after upending global tech. 2026-04-24. https://www.aljazeera.com/economy/2026/4/24/chinas-deepseek-unveils-latest-model-a-year-after-upending-global-tech
• Euronews — China’s DeepSeek releases new AI model V4. 2026-04-24. https://www.euronews.com/next/2026/04/24/chinas-deepseek-releases-new-ai-model-v4-heres-everything-to-know-as-the-ai-race-speeds-up
• VentureBeat — DeepSeek-V4 arrives with near state-of-the-art intelligence at 1/6th the cost of Opus 4.7, GPT-5.5. 2026-04-24. https://venturebeat.com/technology/deepseek-v4-arrives-with-near-state-of-the-art-intelligence-at-1-6th-the-cost-of-opus-4-7-gpt-5-5

Analysis & community

• Artificial Analysis — DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash. 2026-04-24. https://artificialanalysis.ai/articles/deepseek-is-back-among-the-leading-open-weights-models-with-v4-pro-and-v4-flash
• BuildFastWithAI — DeepSeek V4-Pro Review: Benchmarks, Pricing & Architecture. 2026. https://www.buildfastwithai.com/blogs/deepseek-v4-pro-review-2026
• Digital Applied — DeepSeek V4 Launches: 1.6T MoE, 1M Context, 10% KV. 2026. https://www.digitalapplied.com/blog/deepseek-v4-preview-launch-1m-context-efficiency
• DevTk — DeepSeek V4 API Pricing. 2026. https://devtk.ai/en/models/deepseek-v4/
• AI Tool Insight — DeepSeek Unveils V4 at Rock-Bottom Prices With Full Support From Huawei Chips. 2026. https://aitoolinsight.com/deepseek-v4-model-huawei-chips-pricing-open-source/
• BentoML — The Complete Guide to DeepSeek Models: V3, R1, V4 and Beyond. 2026. https://www.bentoml.com/blog/the-complete-guide-to-deepseek-models-from-v3-to-r1-and-beyond
• Knight Li — DeepSeek-V4 Preview Released: 1M Context, Two Models, and API Migration Notes. 2026-04-24. https://www.knightli.com/en/2026/04/24/deepseek-v4-preview-release/
• FelloAI — DeepSeek V4 Released: Everything You Need to Know (April 2026). 2026. https://felloai.com/deepseek-v4/
• LMSYS — DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles. 2026-04-25. https://www.lmsys.org/blog/2026-04-25-deepseek-v4/
• Fortune — DeepSeek unveils V4 model, with rock-bottom prices and close integration with Huawei’s chips. 2026-04-24. https://fortune.com/2026/04/24/deepseek-v4-ai-model-price-performance-china-open-source/
• vLLM Ascend — DeepSeek-V4 deployment tutorial. 2026. https://docs.vllm.ai/projects/ascend/en/v0.13.0/tutorials/DeepSeek-V4.html
• Tao An — DeepSeek’s MODEL1 Leak Reveals V4’s Architectural Blueprint. Medium, 2026. https://tao-hpu.medium.com/deepseeks-model1-leak-reveals-v4-s-architectural-blueprint-28e2bdcc7f37
• Simon Willison — DeepSeek V4 — almost on the frontier, a fraction of the price. 2026-04-24. https://simonwillison.net/2026/Apr/24/deepseek-v4/
• Hacker News — DeepSeek V4 announcement thread. 2026-04-24. https://news.ycombinator.com/item?id=47884971
• Hacker News — DeepSeek V4 tech report thread. 2026-04-24. https://news.ycombinator.com/item?id=47885014
• Hacker News — DeepSeek V4 coding breakdown thread. 2026-04-24. https://news.ycombinator.com/item?id=47885230
• Hugging Face Forums — DeepSeek V4 is live in preview — should your team switch?. 2026. https://discuss.huggingface.co/t/deepseek-v4-is-live-in-preview-should-your-team-switch/175560
• DeepSeek API Docs — JSON Mode guide. https://api-docs.deepseek.com/guides/json_mode. Accessed 2026-04-26.
• DeepSeek API Docs — Function Calling guide. https://api-docs.deepseek.com/guides/function_calling. Accessed 2026-04-26.

Community quantisations and tooling

• Hugging Face — tecaprovn/deepseek-v4-flash-gguf. https://huggingface.co/tecaprovn/deepseek-v4-flash-gguf. Accessed 2026-04-26.
• Hugging Face — mlx-community/deepseek-ai-DeepSeek-V4-Flash-8bit. https://huggingface.co/mlx-community/deepseek-ai-DeepSeek-V4-Flash-8bit. Accessed 2026-04-26.
• All Things How — DeepSeek V4 GGUF Status: What Runs Locally and What Doesn’t. 2026. https://allthings.how/deepseek-v4-gguf-status-what-runs-locally-and-what-doesnt/
• Hugging Face — unsloth/DeepSeek-V4-Pro. https://huggingface.co/unsloth/DeepSeek-V4-Pro. Accessed 2026-04-27.
• Hugging Face — unsloth/DeepSeek-V4-Flash. https://huggingface.co/unsloth/DeepSeek-V4-Flash. Accessed 2026-04-27.
• Dataconomy — Chinese AI Models Hit 61% Market Share On OpenRouter. 2026-02-25. https://dataconomy.com/2026/02/25/chinese-ai-models-hit-61-market-share-on-openrouter/
• AICost — OpenRouter Monthly Token Usage Ranking 2026: Why Chinese Models Like MiniMax M2.5 and Kimi K2.5 Dominate. https://aicost.org/blog/openrouter-monthly-token-usage-ranking-2026-chinese-models-dominate. Accessed 2026-04-27.

Alternative API providers

• OpenRouter — DeepSeek V4 Pro page. https://openrouter.ai/deepseek/deepseek-v4-pro. Accessed 2026-04-27.
• OpenRouter — DeepSeek V4 Flash page. https://openrouter.ai/deepseek/deepseek-v4-flash. Accessed 2026-04-27.
• NVIDIA NIM — DeepSeek V4 Pro reference. https://docs.api.nvidia.com/nim/reference/deepseek-ai-deepseek-v4-pro. Accessed 2026-04-27.
• NVIDIA NIM — DeepSeek V4 Flash reference. https://docs.api.nvidia.com/nim/reference/deepseek-ai-deepseek-v4-flash. Accessed 2026-04-27.
• NVIDIA Developer Blog — Build with DeepSeek V4 Using NVIDIA Blackwell and GPU-Accelerated Endpoints. 2026. https://developer.nvidia.com/blog/build-with-deepseek-v4-using-nvidia-blackwell-and-gpu-accelerated-endpoints/
• NVIDIA NGC Catalog — DeepSeek-V4-Pro container. https://catalog.ngc.nvidia.com/orgs/nim/teams/deepseek-ai/containers/deepseek-v4-pro. Accessed 2026-04-27.
• DeepInfra — DeepSeek-V4-Pro demo. https://deepinfra.com/deepseek-ai/DeepSeek-V4-Pro. Accessed 2026-04-27.

Performance / throughput measurements

• Artificial Analysis — DeepSeek V4 Flash (Max) — Intelligence, Performance & Price Analysis. https://artificialanalysis.ai/models/deepseek-v4-flash. Accessed 2026-04-27.
• BSWEN — What Are the Actual Performance Metrics of DeepSeek V4 Flash for Coding?. 2026-04-26. https://docs.bswen.com/blog/2026-04-26-deepseek-v4-flash-performance-metrics/
• Particula — SGLang vs vLLM in 2026: Benchmarks, Architecture, and When to Use Each. https://particula.tech/blog/sglang-vs-vllm-inference-engine-comparison. Accessed 2026-04-27.
• WaveSpeedAI — DeepSeek V4 Pro vs Flash: Which One for Production?. https://wavespeed.ai/blog/posts/deepseek-v4-pro-vs-flash/. Accessed 2026-04-27.

Adoption + ecosystem (post-launch)

• TechNode — DeepSeek V4 becomes default model for OpenClaw. 2026-04-27. https://technode.com/2026/04/27/deepseek-v4-becomes-default-model-for-openclaw/
• CNN Business — China’s AI upstart DeepSeek drops new model. Will it make waves like last year?. 2026-04-24. https://www.cnn.com/2026/04/24/tech/chinas-ai-deepseek-v4-intl-hnk
• gHacks Tech News — DeepSeek Releases V4 Models With 9.5x Lower Memory Requirements and Huawei Ascend Support. 2026-04-26. https://www.ghacks.net/2026/04/26/deepseek-releases-v4-models-with-9-5x-lower-memory-requirements-and-huawei-ascend-support/
• TechCrunch — DeepSeek previews new AI model that ‘closes the gap’ with frontier models. 2026-04-24. https://techcrunch.com/2026/04/24/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models/
• US News — China’s DeepSeek Rolls Out a Long-Anticipated Update of Its AI Model. 2026-04-24. https://www.usnews.com/news/business/articles/2026-04-24/chinas-deepseek-rolls-out-a-long-anticipated-update-of-its-ai-model

Safety and bias research (V3/R1-era; relevant to V4)

• Adversa AI — AI Red Teaming Reasoning LLM US vs China: Jailbreak Deepseek, Qwen, O1, O3, Claude, Kimi. https://adversa.ai/blog/ai-red-teaming-reasoning-llm-jailbreak-china-deepseek-qwen-kimi/. Accessed 2026-04-27.
• Enkrypt AI — DeepSeek Under Fire: Uncovering Bias & Censorship from 300 Geopolitical Questions. https://www.enkryptai.com/blog/deepseek-under-fire-uncovering-bias-censorship-from-300-geopolitical-questions. Accessed 2026-04-27.
• Promptfoo — 1,156 Questions Censored by DeepSeek. https://www.promptfoo.dev/blog/deepseek-censorship/. Accessed 2026-04-27.
• Promptfoo — CCP-Sensitive-Prompts dataset (1,360 prompts × 68 sensitive topics). https://huggingface.co/datasets/promptfoo/CCP-sensitive-prompts. Accessed 2026-04-27.
• NBC News — On DeepSeek, you can watch AI navigate censorship in real time. https://www.nbcnews.com/tech/innovation/deepseek-censorship-china-rcna189594. Accessed 2026-04-27.