DeepSeek V4 Pro

DeepSeek V4 Pro (also styled DeepSeek-V4-Pro) is a large language model in the DeepSeek series developed by the Chinese AI company DeepSeek. According to Wikipedia's account of the DeepSeek model line, DeepSeek released DeepSeek V4 and V4-Pro on 24 April 2026, followed by DeepSeek-V4.1-Flash on 9 September 2026. The V4 Pro is therefore a member of the V4 generation of DeepSeek models, positioned alongside the standard V4 variant. The DeepSeek model family began in November 2023 with the DeepSeek-LLM series (7B and 67B parameter base and chat models). Subsequent releases included DeepSeek Coder, DeepSeek-MoE, DeepSeek-V2 (which introduced multi-head latent attention, or MLA), DeepSeek-V3, and DeepSeek-R1, the reasoning-focused model whose January 2025 chatbot release became a breakout global success. DeepSeek models are known for their open weights, efficiency-oriented architecture (mixture-of-experts with shared experts, MLA, KV caching), and comparatively low API pricing. The V4 and V4 Pro models continue this lineage. DeepSeek offers its chatbot free through its website and mobile apps, and provides API access to its models on a usage-based pricing model aimed at developers and businesses. The company's models have been described as upending the AI industry and triggering a global AI space race, while also drawing scrutiny over compliance with Chinese government censorship policies and data collection practices in multiple countries. Specific technical details, parameter counts, benchmark results, and pricing for DeepSeek V4 Pro are not provided in the available evidence and are therefore not stated here.

DeepSeek V4 Pro belongs to the DeepSeek series of large language models developed by the Chinese company DeepSeek. The series began in November 2023 with the DeepSeek-LLM models, released in 7B and 67B parameter base and chat forms. These early models used a Llama-style dense decoder-only transformer architecture and were made available under the source-available DeepSeek License. DeepSeek subsequently expanded the family with DeepSeek Coder (a series of eight code-focused models with 16K context), DeepSeek-MoE (16B parameters with 2.7B activated per token), and later DeepSeek-V2, which replaced standard multi-head attention with multi-head latent attention (MLA) to reduce KV cache size and memory usage. DeepSeekMoE introduced shared experts that are always activated alongside routed experts. DeepSeek-V3 and DeepSeek-R1 followed; R1, a reasoning-focused model, became a breakthrough success when its chatbot was released on 10 January 2025 for iOS and Android. By 27 January 2025, DeepSeek-R1 had surpassed ChatGPT as the most-downloaded freeware app on the iOS App Store in the United States, a development linked to an 18% drop in Nvidia's share price. The same day, a large-scale cyberattack disrupted DeepSeek's servers, and the company limited new user registration to mainland China phone numbers, email addresses, or Google account logins. In April 2025, DeepSeek published research with Tsinghua University on DeepSeek-GRM, combining generative reward modeling and self-principled critique tuning, and released DeepSeek-Prover-V2-671B for formal theorem proving and mathematical reasoning. On 24 April 2026, DeepSeek released DeepSeek V4 and V4-Pro, and on 9 September 2026 it released DeepSeek-V4.1-Flash. The V4 Pro is thus a member of the V4 generation, positioned alongside the standard V4 model. The available evidence does not detail the specific architecture, parameter count, or capabilities of V4 Pro beyond its release date and family membership.

• DeepSeek series begins with DeepSeek-LLM models (7B and 67B).

• DeepSeek releases its chatbot for iOS and Android based on DeepSeek-R1.

• DeepSeek-R1 surpasses ChatGPT as the most-downloaded freeware app on the US iOS App Store.

• DeepSeek publishes research on DeepSeek-GRM with Tsinghua University.

• DeepSeek releases DeepSeek-Prover-V2-671B for formal theorem proving.

• DeepSeek releases DeepSeek V4 and V4-Pro.

• DeepSeek releases DeepSeek-V4.1-Flash.

Part of DeepSeek's V4 generation of open-weight, efficiency-focused large language models, offered alongside free chatbot access and low-cost API usage.

• DeepSeek V4 Pro

  Large language model released on 24 April 2026 alongside DeepSeek V4.

• DeepSeek V4

  Standard V4 model released alongside V4-Pro.

• DeepSeek-V4.1-Flash

  Model released on 9 September 2026 following V4 and V4-Pro.

1. 1DeepSeek (AI) - Wikipediaen.wikipedia.org