DeepSeek models
DeepSeek's open-weight model line, from V3 and the market-moving R1 reasoning model to the 552B-parameter V4 generation.
DeepSeek’s models are known for frontier-competitive quality at far lower cost.
| Model | Released | Notes |
|---|---|---|
| DeepSeek-V3 | Dec 2024 | 671B MoE; low reported training cost |
| DeepSeek-R1 | Jan 2025 | Open reasoning model; triggered a historic one-day Nvidia selloff |
| DeepSeek V4 Pro | 2026 | Retired: from Sep 14 2026 requests to V4-Pro are served by V4.1-Flash until a V4.1-Pro arrives |
| DeepSeek V4.1 Flash | Sep 10 2026 | 552B parameters (8B active reading input, 16B writing), 1M context, MIT weights on Hugging Face; used in the ARTEX bank breach |
| V4.1 Flash-1017 | Oct 17 2026 | Minor revision under the same API name with the same rate card (secondary coverage; Search & Fetch tool pricing reported from Oct 31) |
Microsoft AI benchmarks its MAI-Base-2 pre-training efficiency against DeepSeek-V4-Pro-Base. The V4.1-Flash serving stack, DeepInfer, is open source with Nvidia and Huawei Ascend kernels.
Timeline
Every dated event in the wiki that involves DeepSeek models, newest first.
- Operation Snowball: jailbroken DeepSeek agent breaches a European telecom via a Cisco BroadWorks flaw
Malbear Labs found an exposed workspace showing a Chinese-speaking operator running a custom agent on DeepSeek Flash under a ~78,000-character jailbreak. It entered via CVE-2026-20320, a Cisco (Talos) BroadWorks XXE fixed in August, then used old Liferay and JBoss flaws. Victims: a telecom and 13 organizations on its infrastructure; data claims are unverified. CISA added the CVE to KEV on Nov 10.
- DeepSeek open-sources DeepInfer
Apache-2.0 serving stack for V4.1-Flash with kernels for Nvidia SM100 and Huawei Ascend 950.
Sources: GitHub
- DeepSeek releases V4.1 Flash-1017
Minor revision of the 552B V4.1 Flash (released Sep 10) under MIT with 1M context, same API name and rate card; Search & Fetch tools reported as separately priced.
Full article: DeepSeek modelsSources: The Next Web, “DeepSeek launches V4.1-Flash and retires V4-Pro, its flagship model” (for the base V4.1 Flash release); Hugging Face
- AI agents mass-exploit PaperCut zero-days, compromising 440 servers (PaperCut AI agent campaign)
A likely Russian-speaking actor used AI to build and test exploits for PaperCut NG/MF flaws CVE-2026-81578 and CVE-2026-82078 (disclosed Aug 27), then ran hundreds of agents on OpenAI's Codex harness with a DeepSeek model against at least 440 instances at 395 organizations in 48 countries, per GreyNoise. It went from an empty workspace to first remote code execution in under four hours and reached domain admin at 12 organizations.
Connections
DeepSeek models and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →
What links here
Incident ARTEX breach of Korean banks · Operation Snowball · PaperCut AI agent campaign
Concept Context window · Open-weight vs. closed models
AI lab DeepSeek · Microsoft AI (MAI)
Organization Malbear Labs