DeepSeek

DeepSeek is a Hangzhou AI company that develops the DeepSeek language models, including the V4 and R1 series, released as open weights and through its API and chat app.

https://www.deepseek.com →

News

researchDeepSeek

DeepSeek open-sources Super-agent prototype: Router-Worker design hits 15.7-minute P95 in simulation

DeepSeek has published an engineering write-up and a GitHub prototype for Super-agent, a two-layer agent architecture built around a Router and parallel Workers. In a simulated environment of about 300 messages and 20 work requests, DeepSeek reports a median task completion time of 12.4 minutes and a 95th-percentile time of 15.7 minutes.

3 min read
product updateDeepSeek

DeepSeek website code references Cyber, a cybersecurity model with identity-gated API access, in limited testing

Developers found references to a cybersecurity model called Cyber in the front-end code of DeepSeek's website, with two model IDs, cyber-personal-v3 and cyber-enterprise-v4, and a key system gated by real-name verification. DeepSeek has not announced the model, and launch, access and pricing remain unclear.

2 min read
model releaseDeepSeek

DeepSeek V4.1 Flash 1017: $0.15/$0.6 per 1M tokens, 1M context, 67% hallucination rate on non-answers

DeepSeek has upgraded its open-weight V4.1 Flash to revision 1017, a 552B-parameter MoE model with 16B active parameters and a 1 million token context window. Off-peak API pricing is $0.15 per 1M input tokens and $0.6 per 1M output tokens. Artificial Analysis data shows strong intelligence scores but weak results on hallucination.

3 min read
model releaseDeepSeek

DeepSeek Releases V4.1-Flash: 552B MoE Model Cuts KV Cache to 890 Bytes Per Token

DeepSeek has released V4.1-Flash, a 552B-parameter multimodal Mixture-of-Experts model supporting 1M-token context and activating only 8B parameters during prefill. The model uses a new Causal Encoder-Decoder architecture and Compressed Sparse Attention 2 to cut global KV cache to 890 bytes per token, roughly a quarter of its predecessor.

3 min read
changelogDeepSeek

DeepSeek V4 Flash 'O731' Nearly Matches GPT-5.6 Luna, Costs 60% Less to Run

DeepSeek has updated its budget model V4 Flash to version '0731,' pushing its Artificial Analysis Intelligence Index score to 50 — just one point behind OpenAI's GPT-5.6 Luna — while costing an estimated 60 percent less per task. The MIT-licensed model keeps its 284B-parameter architecture but shows major gains in agentic benchmarks and token efficiency.

2 min read