{"passport":{"slug":"deepseek-r1","display_name":"DeepSeek R1","model_type":"language","parameter_count":"671B total (37B active per token)","context_window":"128k tokens","claimed_developer":"DeepSeek","confirmed_developer":"DeepSeek","developer_confirmed_at":"2025-01-20T00:00:00+00:00","developer_source":"Official DeepSeek release — api-docs.deepseek.com + GitHub + HuggingFace","first_appeared_at":"2025-01-20T00:00:00+00:00","first_appeared_on":"github","status":"active","availability":null,"deployment_status":"not_deployed","availability_scope":null,"gate":3,"eu_ai_act":"unregistered","regulatory_status":{"us":"available (app restricted in some contexts)","china":"compliant — enforces CCP content policies","eu_ai_act":"unregistered","gdpr_notes":"Chinese developer, subject to Chinese data laws and CCP ideology censorship requirements"},"open_source":true,"weights_available":true,"license":null,"provenance_chain":[{"role":"developer","entity":"DeepSeek","source":"official release","confirmed_at":"2025-01-20"},{"role":"parent company / funder","entity":"High-Flyer (hedge fund)","source":"public reporting + Wikipedia","confirmed_at":"2025-01-20"},{"role":"founder and CEO of both DeepSeek and High-Flyer","entity":"Liang Wenfeng","source":"public reporting","confirmed_at":"2025-01-20"}],"parent_model":"deepseek-v3","model_family":"DeepSeek","superseded_by":null,"supersedes":[],"recommended_replacement":null,"product_links":null},"events":[{"event_type":"update_released","event_date":"2025-05-28T00:00:00+00:00","title":"DeepSeek R1-0528 released — adds system prompts, JSON output, function calling","detail":"DeepSeek releases updated version R1-0528 under MIT license. Adds system prompt support, JSON output, and function calling — making it significantly more suitable for agentic AI use cases. Reduced hallucination rates claimed. Average reasoning depth increases from 12,000 to 23,000 tokens per question.","source_url":"https://en.wikipedia.org/wiki/DeepSeek","source_name":"Wikipedia / DeepSeek official"},{"event_type":"regulatory_event","event_date":"2025-01-28T00:00:00+00:00","title":"CCP ideology enforcement confirmed in model responses","detail":"Researchers and journalists document DeepSeek R1 enforcing Chinese Communist Party content policies. Model initially describes Taiwan as 'an inalienable part of China’s territory.' Responses on Tiananmen Square, Hong Kong, and Xinjiang are filtered or censored. Locally hosted instances occasionally produce CCP-aligned responses. Western researchers find workarounds using letter substitution.","source_url":"https://en.wikipedia.org/wiki/DeepSeek_(chatbot)","source_name":"NBC News / Wikipedia / multiple researchers"},{"event_type":"deployment_spotted","event_date":"2025-01-27T00:00:00+00:00","title":"DeepSeek surpasses ChatGPT on iOS — Nvidia loses $600B in market cap","detail":"DeepSeek R1 becomes the most downloaded free app on the US iOS App Store, surpassing ChatGPT. Simultaneously, Nvidia stock falls 17-18% — approximately $600 billion in market capitalisation lost in a single day. Marc Andreessen calls it 'AI Sputnik moment.' Nasdaq Composite drops 3.4% at market open. Triggers US government concern about export control effectiveness.","source_url":"https://en.wikipedia.org/wiki/DeepSeek_(chatbot)","source_name":"Wikipedia / multiple confirmed sources"},{"event_type":"security_concern","event_date":"2025-01-27T00:00:00+00:00","title":"Large-scale cyberattack forces DeepSeek to limit new registrations","detail":"DeepSeek suffers a large-scale cyberattack on the same day it hits #1 on the iOS App Store. New user registration temporarily limited to phone numbers from mainland China, email addresses, or Google account logins. Service disruptions reported globally.","source_url":"https://en.wikipedia.org/wiki/DeepSeek_(chatbot)","source_name":"Multiple news sources / Wikipedia"},{"event_type":"first_appearance","event_date":"2025-01-20T00:00:00+00:00","title":"DeepSeek R1 released — open source under MIT license","detail":"DeepSeek releases R1 and R1-Zero (671B parameters each) plus distill variants ranging from 1.5B to 70B parameters. MIT license. Performance on par with OpenAI o1 on reasoning benchmarks including AIME and MATH. Training cost claimed at fraction of competing models. Model and app released simultaneously.","source_url":"https://api-docs.deepseek.com/news/news250120","source_name":"DeepSeek official API documentation"}],"delta":{"days_unattributed":0,"claimed_developer":"DeepSeek","confirmed_developer":"DeepSeek","match":true}}