DeepSeek-V3
DeepSeek-V3 is an open-weight mixture-of-experts language model from DeepSeek, announced on December 26, 2024, with 671B total and 37B activated parameters, trained on 14.8T tokens.
Retired. DeepSeek’s API no longer serves DeepSeek-V3: its change log records deepseek-chat moving to DeepSeek-V3-0324 on March 24, 2025 and to later versions after that, and its pricing page now lists only DeepSeek-V4 models. The weights remain on Hugging Face.
Released
December 26, 2024
Type
llm
License
open-weight
DeepSeek Model License (commercial use allowed); code under MIT
Context
128,000 tokens
Capabilities
Architecture
Parameters: 671B total, 37B active (mixture of experts)
Sources
Every figure on this page was checked against these pages on . A figure we could not source is left out.
- DeepSeek: DeepSeek-V3 release (opens in a new tab)
- Model card: DeepSeek-V3 (opens in a new tab)
- DeepSeek API pricing (opens in a new tab)
- DeepSeek API docs: Change log (opens in a new tab)
Spotted something out of date? Tell us.
Links
In the news: DeepSeek-V3 and DeepSeek
MarkTechPost
Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads
Reflection AI has introduced Beam, its inaugural open-weight model, on March 18, 2024. Beam is a sparse Mixture-of-Experts (MoE) architecture boasting a total of 501 billion parameters, with 23 billion active parameters per token. This model is specifically engineered for coding, reasoning, and agentic workloads. According to the Reflection AI team, Beam is positioned to directly compete with larger open models such as GLM 5.2, while demonstrating a significant advantage in inference compute efficiency, using 3 to 4 times less compute on reasoning benchmarks. Currently, Beam is not available for self-hosting but is undergoing final red-teaming. Early access is being managed through a waitlist on the Reflection platform. Beam functions as a general agent model, developed from the ground up by Reflection AI, with a primary focus on enterprise-level coding and agentic applications. Reflection AI frames Beam as a significant advancement in the Western open-weight model landscape. The research team acknowledges that while Kimi K3 may still lead in raw capability, Beam's key selling proposition is its efficiency during inference. Users can adjust a "reasoning effort" parameter, where lower settings yield concise answers and higher settings enable more extensive reasoning for complex tasks. This allows teams to tailor the model's effort to the specific difficulty of a task and their available compute budget. The pretraining of Beam involved an extensive dataset of 23.8 trillion tokens, sourced from the web, public repositories, and proprietary licensed datasets. The Reflection AI team reported that their curation process removed approximately 95% of raw internet tokens, while retaining about 1.8 trillion high-quality tokens that might have been discarded by conventional filtering methods. The model's architecture integrates local and global attention mechanisms with finely routed experts. Its load balancing strategy builds upon DeepSeek-V3's auxiliary-loss-free balancing, incorporating a cosine decay for expert-bias updates. This approach ensured that the most utilized expert experienced a load of only 1.04 times the average by the conclusion of pretraining. Throughout the 52 layers, residual norms were maintained within bounds through depth-based scaling, SandwichNorm, attention gating, and FP32 residual accumulation. The entire pretraining process was completed in under four weeks, utilizing 6,144 NVIDIA GB300 NVL72 GPUs. The reported goodput reached 92.3% near the end of the training period, with nine semi-automatic rewinds implemented. During mid-training, the effective context window was extended to 1 million tokens.
MarkTechPost
DeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness
DeepSeek has released an official desktop application for its open-source agent harness, DeepSeek Harness (dsh), with the v0.2 preview. Installers are available for macOS, specifically for Apple silicon, and for Windows 64-bit operating systems. This preview version is deployable immediately and can be downloaded from deepseek.com/harness or accessed via the command line using 'npx @deepseek-ai/dsh web'. DeepSeek has cautioned that compatibility-breaking changes are expected in future releases. The v0.2 preview enhances the harness's capabilities to target both everyday work and coding tasks. Key built-in features include preloaded, commonly used tools for office productivity and software development. A new plugin management interface allows users to install, configure, enable, and disable plugins through a dedicated page. For code and file review, a right-hand sidebar provides previews of generated files and code changes. The harness now supports document work, enabling users to input documents, spreadsheets, or PDFs and request the generation of charts or slides. An automation task plugin has been introduced, which schedules recurring prompts and includes a run history with editable frequency settings. The desktop build also integrates the dsh command-line interface, eliminating the need for separate Node.js or pnpm installations. Users can authenticate using a DeepSeek account or by providing an API key. According to the release candidate 1 (rc.1) notes, account-based models gain access to web search functionality without requiring an additional API key. The architecture of DeepSeek Harness is designed around a plugin-centric approach, where the model adapter, tool registry, and agent loop are all implemented as plugins, allowing for modularity and replacement of any layer. DeepSeek first open-sourced dsh in August 2026 under the MIT license. The harness runs on Cordis, as detailed in the research paper ‘A Programming Paradigm for Spatiotemporal Composability’. The harness is not restricted to DeepSeek's own models, with a provider guide available for integrating third-party providers and custom OpenAI-compatible endpoints. The project's GitHub repository has garnered significant attention, exceeding 240,000 stars and 24,000 forks. DeepSeek Harness also serves as DeepSeek's internal evaluation framework. The API changelog indicates that Code Agent benchmarks, such as the Terminal Bench 2.1, have been run using DeepSeek Harness in a minimal configuration. For instance, the DeepSeek-V4-Flash-0731 model achieved a score of 82.7 on Terminal Bench 2.1 within this setup. The harness supports four distinct session modes.
Decrypt
China Probes DeepSeek and Moonshot Over Alleged Data Leaks to Anthropic's Claude
China's Cyberspace Administration (CAC) has launched an investigation into two prominent artificial intelligence companies, DeepSeek and Moonshot AI, following accusations of data mishandling. The probe was initiated after Anthropic, a leading AI research firm, alleged that both DeepSeek and Moonshot AI had secretly channeled millions of user interactions through Anthropic's Claude models. The purpose of this alleged data transfer was reportedly to train their own proprietary AI models without explicit user consent or proper authorization. This investigation highlights growing concerns within China regarding data privacy and the ethical implications of AI development. The CAC, as the primary internet regulator in China, is tasked with overseeing online content, data security, and the operations of internet companies within the country. The agency's involvement signals the seriousness with which Beijing is treating these allegations. DeepSeek, known for its open-source large language models (LLMs), and Moonshot AI, a competitor in the AI space, are significant players in China's rapidly expanding AI industry. Their alleged actions, if proven, could have substantial repercussions for their operations and reputation. Anthropic's accusation suggests a sophisticated method of data exfiltration, where user data, potentially including sensitive information exchanged with Claude, was rerouted for external model training. This practice, if confirmed, would violate data privacy regulations and potentially breach user trust. The scale of the alleged data leak, involving "millions of user exchanges," underscores the potential impact on a vast number of individuals whose data may have been compromised. The CAC's investigation will likely involve examining the technical infrastructure and data handling practices of both DeepSeek and Moonshot AI to ascertain the veracity of Anthropic's claims and determine the extent of any wrongdoing. The outcome of this investigation could set a precedent for AI development and data governance in China. Regulators may impose stricter guidelines on data usage for AI training, enhance oversight mechanisms, and enforce penalties for non-compliance. This situation also underscores the competitive landscape within the global AI sector, where companies are constantly seeking advantages through data and model improvement. The alleged data leak raises questions about the security protocols and ethical boundaries that govern the development and deployment of advanced AI technologies. The CAC's findings will be closely watched by both domestic and international stakeholders in the AI industry.
Decrypt
UN Security Council Will Get Advice on AI Risks From Tech Giants Building It
A group of leading artificial intelligence developers is set to brief the United Nations Security Council this week on the potential risks associated with advanced AI technologies. Among the participants scheduled to address the council is DeepSeek, a prominent AI research organization. DeepSeek will share its perspectives on the challenges and dangers posed by rapidly evolving AI systems. Also slated to speak is Dario Amodei, the CEO of Anthropic, a company known for its focus on AI safety and ethical development. Amodei has previously voiced significant concerns regarding the geopolitical implications of AI, notably identifying China's government as a major threat to the industry's responsible advancement over the past year. The Security Council's engagement with these tech leaders signifies a growing recognition within international governance bodies of the need for expert input on the complex and multifaceted risks of artificial intelligence. This briefing is part of a broader effort by the UN to understand and potentially regulate AI. The Security Council, typically focused on matters of international peace and security, is increasingly grappling with how emerging technologies like AI could impact global stability, warfare, and societal structures. The inclusion of major AI developers in these discussions highlights a shift towards collaborative policymaking, where the creators of the technology are directly consulted by international bodies. This approach aims to leverage the deep technical knowledge of these companies while also ensuring that their development trajectories align with global safety and ethical standards. The discussions are expected to cover a range of AI-related risks, including potential misuse in cyber warfare, the spread of disinformation, economic disruption, and the existential threats that some researchers believe advanced AI could pose. The participation of companies like DeepSeek and Anthropic underscores the critical role that industry players have in shaping the future of AI. DeepSeek, known for its open-source models and research into large language models, brings a perspective from a significant player in the AI development landscape. Anthropic, founded by former OpenAI researchers, has positioned itself as a leader in developing AI systems that are helpful, honest, and harmless, with a strong emphasis on safety research and alignment. Dario Amodei's past statements about China's government suggest that discussions may touch upon international competition in AI development and the potential for AI to be used in ways that challenge existing international norms and security frameworks. The UN Security Council's decision to convene these discussions reflects the urgent need for a global dialogue on AI governance, moving beyond national interests to address a technology with universal implications. This engagement by the UN Security Council with AI developers is a significant step in the ongoing global conversation about AI governance. It acknowledges that effective regulation and risk mitigation require a deep understanding of the technology itself, which is best provided by those at the forefront of its creation. The insights shared by DeepSeek and Dario Amodei are expected to inform the council's deliberations on how to foster innovation while simultaneously safeguarding against the potential downsides of advanced AI. The outcomes of these discussions could influence future international policies and agreements related to AI development and deployment, aiming to ensure that this powerful technology benefits humanity as a whole.
The Hacker News
Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Claude Distillation Attacks
Anthropic announced on Thursday that it has identified and disrupted industrial-scale illicit distillation attacks targeting its Claude AI model. These sophisticated attacks were traced back to seven distinct artificial intelligence laboratories based in China. The identified entities include prominent organizations such as Alibaba, Moonshot, DeepSeek, Z.ai (also known as Zhipu), and MiniMax. The attacks involved knowledge distillation, a legitimate machine learning technique where a larger, more capable AI model acts as a 'teacher' to train a smaller model. However, in this context, the labs were allegedly using Anthropic's powerful Claude models to illicitly train their own AI systems without authorization, bypassing Anthropic's terms of service and potentially infringing on intellectual property rights. Anthropic's security team detected these activities by monitoring for unusual patterns of access and data exfiltration that were indicative of large-scale, systematic attempts to replicate Claude's capabilities. The company stated that these distillation efforts were "industrial-scale," suggesting a significant investment of resources and computational power by the implicated labs. Such attacks aim to extract proprietary knowledge and model parameters from a teacher model to create a student model that mimics its performance, often at a fraction of the training cost and time. By engaging in this unauthorized distillation, the Chinese labs sought to gain a competitive advantage by rapidly developing advanced AI models based on Anthropic's research and development. In response to the discovery, Anthropic took immediate action to mitigate the threat. The company stated that it has taken steps to block the identified actors and prevent further unauthorized access and misuse of its Claude models. This disruption aims to protect Anthropic's intellectual property and maintain the integrity of its AI systems. The company emphasized that knowledge distillation is a valid technique when conducted ethically and with proper licensing, but the actions of these seven labs constituted a violation of their acceptable use policies. The incident highlights the ongoing challenges in securing advanced AI models against sophisticated intellectual property theft and unauthorized replication attempts, particularly in a competitive global AI landscape. Anthropic has not disclosed the specific versions of Claude that were targeted in these attacks, nor the exact methods used by the labs to extract the necessary data for distillation. However, the scale and systematic nature of the attacks suggest a coordinated effort. The company's proactive stance in identifying and reporting these activities underscores the increasing importance of AI security and the need for robust defenses against emerging threats. The involvement of multiple well-known Chinese AI companies also points to the intense competition and rapid development occurring within China's AI sector, where the pursuit of cutting-edge capabilities may sometimes lead to questionable practices. Anthropic's disclosure serves as a warning to other AI developers and a signal of the evolving security landscape in artificial intelligence.