Releases

DeepSeek retired the experimental V4 Flash Vision model and routed it to V4.1 Flash.

DeepSeek retired V4 Flash and temporarily routed its model name to V4.1 Flash.

DeepSeek released a smaller native-multimodal V4.1 architecture model with upgraded coding, reasoning and agents.

DeepSeek reduced API prices following the V4.1 Flash release.

DeepSeek reportedly moved toward a potential mainland listing while its next financing remained in progress.

DeepSeek released an updated agent harness with support for V4 Flash Vision.

DeepSeek launched an experimental native multimodal V4 Flash model.

DeepSeek added native Responses API support and low, high and max thinking effort controls.

DeepSeek rolled out the GA V4 Pro model across app, web and API with stronger production agent capabilities.

DeepSeek reportedly resumed a second round seeking close to $8B; it remained unclosed.

DeepSeek released the official Flash API in public beta with upgraded coding and agent performance.

Reuters reported DeepSeek was planning a second fundraising round ahead of a potential mainland IPO. The target valuation is retained as metadata because the round was not completed.

A Chinese stock-exchange filing disclosed an indirect 0.8265% stake and implied a RMB350.88B ($51.82B) company valuation.

DeepSeek reportedly raised more than RMB50B ($7.4B) in a founder-controlled structure at a valuation exceeding $50B.

Reuters reported DeepSeek was set to raise about RMB50B from investors including Tencent and CATL, at a projected $52B-$59B post-money valuation.

The efficient V4 Flash preview became available through DeepSeek's API.

DeepSeek introduced the V4 family through Pro and Flash preview API models.

The V4 Pro preview became available through OpenAI- and Anthropic-compatible interfaces.

DeepSeek published its second-generation visual causal-flow OCR model.

DeepSeek offered a temporary high-compute reasoning endpoint until December 15.

DeepSeek upgraded chat and reasoner services to the production V3.2 model.

DeepSeek released its self-verifiable mathematical reasoning model.

DeepSeek introduced its open OCR and document-understanding research family based on visual-text compression.

DeepSeek published its optical-compression model for OCR and document understanding.

DeepSeek released the experimental V3.2 model with sparse attention.

DeepSeek improved language consistency and its code and search agents.

DeepSeek released a hybrid thinking model with improved reasoning efficiency and agent capabilities.

DeepSeek upgraded R1 with stronger reasoning, reduced hallucination and JSON/function-calling support.

DeepSeek released 7B and 671B theorem-proving models and ProverBench.

DeepSeek upgraded V3 with stronger reasoning, coding, writing, search and function calling.

South Korea's privacy regulator suspended new downloads while reviewing personal-data protections.

Australia directed federal entities to remove and prevent DeepSeek products on government systems and devices.

Italy's data-protection authority ordered an immediate restriction on DeepSeek's processing of Italian users' personal data.

DeepSeek released scaled 1B and 7B Janus-Pro checkpoints with improved understanding and image generation.

DeepSeek's assistant reached number one among free iPhone applications in the United States as adoption surged.

DeepSeek temporarily limited new registrations following a large-scale malicious cyberattack.

DeepSeek released the Llama 70B-based distilled R1 reasoning checkpoint.

DeepSeek released the Llama 8B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 14B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 1.5B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 7B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 32B-based distilled R1 reasoning checkpoint.

DeepSeek published R1 under the MIT License and made it available on web and API.

DeepSeek published its pure reinforcement-learning reasoning checkpoint.

DeepSeek introduced its reinforcement-learning reasoning model family and technical report.

DeepSeek released its consumer assistant for iOS and Android.

DeepSeek released DeepSeek-V3 weights and upgraded the deepseek-chat API.

DeepSeek introduced its 671B-parameter V3 mixture-of-experts architecture.

DeepSeek published its second-generation mixture-of-experts vision-language models.

The final V2.5 update improved mathematics, coding, writing, file handling and web-page summarization.

DeepSeek released JanusFlow using rectified flow for image generation.

DeepSeek released Janus 1.3B for multimodal understanding and image generation.

DeepSeek introduced the Janus family for unified multimodal understanding and generation.

DeepSeek merged its general and coding capabilities into V2.5 and published weights.

DeepSeek released an improved formal theorem prover using reinforcement learning and search.

DeepSeek introduced disk-based context caching to reduce repeated-token costs.

The API gained JSON mode, function calling, chat-prefix completion and fill-in-the-middle completion.

DeepSeek upgraded the deepseek-coder API to the July Coder V2 checkpoint.

DeepSeek updated its chat model with stronger code, math, reasoning and role-play performance.

DeepSeek released 16B and 236B mixture-of-experts coding models with 128K context.

DeepSeek upgraded its coding API to the Coder V2 generation.

DeepSeek published its first open theorem-proving model for Lean 4.

DeepSeek introduced its formal theorem-proving model research line.

DeepSeek upgraded deepseek-chat with improved instruction following, JSON output and RAG performance.

DeepSeek opened an OpenAI-compatible API for its chat and coding models.

DeepSeek published DeepSeek-V2 base and chat weights with 128K context.

DeepSeek introduced its efficient second-generation mixture-of-experts architecture.

DeepSeek introduced an open vision-language model family for real-world multimodal understanding.

DeepSeek released 1.3B and 7B base and chat checkpoints.

Base, instruction and reinforcement-learning DeepSeekMath 7B checkpoints were published.

DeepSeek introduced its open mathematical reasoning research line.

DeepSeek released an updated 7B coding model that later served as the initialization for DeepSeekMath.

DeepSeek LLM 67B Base became publicly available as part of DeepSeek's first general LLM release.

DeepSeek published its first general-purpose 7B and 67B open language models.

DeepSeek LLM 7B Base became publicly available as part of DeepSeek's first general LLM release.

DeepSeek LLM 67B Chat became publicly available as part of DeepSeek's first general LLM release.

DeepSeek LLM 7B Chat became publicly available as part of DeepSeek's first general LLM release.

Base and instruction-tuned DeepSeek Coder checkpoints were published for code completion and generation.

DeepSeek released its first open code-model family, spanning 1B to 33B parameters and trained on two trillion tokens.

DeepSeek launched its web conversational assistant.

DeepSeek began offering web-based access to its models through the DeepSeek Chat service.

Hangzhou DeepSeek Artificial Intelligence Co., Ltd. was established by Liang Wenfeng as an AGI-focused research company originating from High-Flyer.