{"city":"Hangzhou","name":"DeepSeek","slug":"deepseek","facts":[{"date":"2026-09-20","type":"revenue_band","value":"<100m","source":"rsshub.bkraju.com — 2026-09-20"},{"date":"2026-09-20","type":"product_launch","value":"Released V4.1-Flash model with 1 million-token context and MoE architecture","source":"rsshub.bkraju.com — 2026-09-20"},{"date":"2026-09-14","type":"product_launch","value":"Began limited test of AI voice interaction with four voice profiles","source":"TechNode"},{"date":"2026-09-11","type":"product_launch","value":"Added a 196-billion-parameter N-gram conditional memory module to V4.1-Flash","source":"The Register"},{"date":"2026-09-11","type":"product_launch","value":"Launched V4.1-Flash with 763 billion parameters and substantially leaner serving architecture","source":"The Register"},{"date":"2026-09-11","type":"product_launch","value":"Released V4.1-Flash with 763 billion parameters and lean-serving architecture","source":"The Register"}],"peers":[],"stage":null,"state":"Zhejiang","aliases":["DeepInfer","DeepSeek Harness","DeepSeek-V4.1-Flash","DeepSeek-V4.1-Flash-1017","High-Flyer","Super-agent","V4.1-Flash-1017","深度求索"],"country":"China","domains":[{"why":"DeepSeek develops large language and multimodal foundation models, including the V4.1-Flash family, released as open-source and open-weight models.","list":null,"rank":null,"tier":"member","layer":null,"total":77,"domain":"AI Core","scores":{"gate_ne":["firepower"],"leverage":8,"position":8,"progress":10,"firepower":8,"commercial":4,"deployment":10,"production":6,"technology":6},"edition":null,"segment":null}],"founded":null,"signals":[{"date":"2026-11-01","header":"DeepSeek unveils experimental architecture for a super agent that coordinates multiple AI assistants","outlet":"Interesting Engineering"},{"date":"2026-11-01","header":"DeepSeek released a new Agent architecture","outlet":"Yicai"},{"date":"2026-11-01","header":"DeepSeek previews a 'super agent' for autonomous multi-task AI workflows","outlet":"SCMP Tech"},{"date":"2026-11-01","header":"DeepSeek builds a super-agent for autonomous coding","outlet":"Tech in Asia"},{"date":"2026-11-01","header":"DeepSeek Harness team open-sources an MIT-licensed Super-agent prototype that finds and fixes bugs on its own","outlet":"Pandaily"},{"date":"2026-10-31","header":"DeepSeek Super-agent uses dual-layer, event-driven design with storage-compute separation and an external Watchdog","outlet":"Yicai"}],"website":"deepseek.com","category":"AI · Models & applications","categories":[{"name":"Open-weight labs","domain":"AI Core","parent":"AI · Models & applications"}],"last_round":null,"updated_at":"2026-11-02T16:32:13.247+00:00","facts_total":117,"positioning":"China's most influential lab, historically a renter of compute funded by its parent hedge fund; now building its first owned gigawatt site on Huawei Ascend for inference while training remains on Nvidia.","sub_category":"Open-weight labs","signals_total":464,"url":"https://www.frontiertechintelligence.com/companies/deepseek","title":"DeepSeek · ai core, Hangzhou, Zhejiang, China · scores, facts and signals · Frontier Tech Intelligence","description":"DeepSeek develops large language and multimodal foundation models, including the V4.1-Flash family, released as open-source and open-weight models. AI Core member (score 77) 464 signals and 117 facts with sources.","publisher":"Frontier Tech Intelligence (Future Frontiers, a trade name of Essential Cognition Inc.)","note":"Assessment by Frontier Tech Intelligence; facts and signals carry their sources."}