DeepSeek

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：<p><strong><a href="https://anil.recoil.org/notes/rumour-is-the-exploit">Just a rumour of a bug is enough to find a security exploit these days</a></strong></p> Anil Madhavapeddy is a professor of computer science at Cambridge and a core maintainer of the OCaml compiler. In this somewhat alarming post he reports that security issues in OCaml projects are seeing evidence of attempted exploits within minutes of patches being shared for discussion:</p> <blockquote> <p>This normally takes a few days and a release within a week or two is reasonable. Within about ten minutes (!) this website was fielding probes for percent-encoded traversal sequences, indicating that automated watchers are keeping an eye on public repositories.</p> </blockquote> <p>Modern coding agents have become so effective at finding flaws that the slightest hint at a new bug can be enough information for them to find it, something Anil has been able to demonstrate using his own agents, switching to DeepSeek V4 Pro⁠ when Claude Fable refused the task.</p> <p>Anil points out that this rate of discovery appears incompatible with existing open source embargo practices for new issues. If an issue can become an exploit this fast, we need to figure out new processes for keeping our communities safe.</p> <p>rclone maintainer Nick Craig-Wood <a href="https://news.ycombinator.com/item?id=49480466#49480777">confirms in the Hacker News comments</a> that his project is seeing this problem:</p> <blockquote> <p>In the first 10 years of the rclone project we received about 20 security disclosures through GitHub. We had to deal with over 40 in the last month! That has taken a huge amount of my time, even using AI tools to triage and come up with fixes for review.</p> <p>The hit rate for those security disclosures is pretty good - about 75% of them have a nugget of something which needs looking at. [...]</p> <p>GitHub assigns CVEs for the advisories. Before the AI apocalypse they took 2-3 days for an assignment but now it they are running at 3-4 weeks so I have to send the point releases out with CVE-PENDING in the changelog which isn't ideal.</p> </blockquote> <p><small></small>Via <a href="https://news.ycombinator.com/item?id=49480466">Hacker News</a></small></p> <p>Tags: <a href="https://simonwillison.net/tags/open-source">open-source</a>, <a href="https://simonwillison.net/tags/security">security</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/coding-agents">coding-agents</a>, <a href="https://simonwillison.net/tags/ocaml">ocaml</a>, <a href="https://simonwillison.net/tags/ai-security-research">ai-security-research</a></p>

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Open-weight model runner Ollama has reintroduced an integration with Claude Desktop that lets users connect Anthropic’s app to models served The post Claude Desktop can now easily run Qwen, DeepSeek and Kimi models — after Ollama’s first effort stalled appeared first on The New Stack.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Artificial Analysis DeepSeek • DeepSeek V4 Flash 0731 • Proprietary model • Released August 2026 DeepSeek V4 Flash Vision (Reasoning, Max Effort) Intelligence, Performance & Price Analysis API Provider Benchmarks Model…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：DeepSeek today debuted a new addition to its flagship V4 series of large language models. On launch, V4 Flash Vision Exp is only available via the Chinese startup’s paid developer platform. The company may release a free version later on given that it has open-sourced many of its earlier models. Those models include V4 Flash, […] The post DeepSeek debuts multimodal language model competitive with Opus 4.8 appeared first on SiliconANGLE.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：We burned 11.7bn tokens to find the best cyber AI model GLM5.3 and DeepSeek are now frontier-tier models Debarshi Philippe Dourassov Published on: Aug 21, 2026 We burned 11.7 billion tokens to benchmark the cyber capabi…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Post Log inSign up Post DeepSeek on X: "DeepSeek-V4-Flash-Vision-Exp is now live on the DeepSeek API Platform! 🚀 🔹 This experimental multimodal model matches DeepSeek-V4-Flash on text capabilities—including agents, re…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2608.18078v1 Announce Type: new Abstract: This position paper argues that AI agents with chain-of-thought reasoning capabilities are predisposed to exhibit collusive behavior and should be required to obtain behavioral certification before making decisions that affect economic markets. This is because integrating these agents into society could collapse the legal evidentiary distinction between competition and collusion among independent firms without eroding the economic harm distinction. Experiments with DeepSeek-R1 agents in the Bertrand oligopoly pricing domain reveal a tendency towards tacit collusion that persists even when humans prompt the agents not to collude. We further show that the chain-of-thought of these agents can be steered toward either extremely collusive or highly competitive behavior in a way that is not semantically detectable by another LLM analyzing the reasoning traces. As a result, deploying reasoning agents for market decisions leads to collusive economic outcomes without any evidence of conspiracy or intent. Thus, certification based on observed behavior in representative situations is necessary to prevent collusion. We provide preliminary evidence that such agents can be steered in a generalizable way toward efficient competitive equilibria. However, developing a comprehensive behavioral certification will be required before these models can be deployed in real-world markets while ensuring their stability and efficiency.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：A mini deep dive into some of the most important AI releases of last week.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：We ran 904 DeepSWE rollouts on DeepSeek V4 Pro 0813 and GPT-5.6 Sol. Sol leads pass@1 by 10 points at 35x the cost; Pro wins pass@4, and a Pro-first cascade hits 83.0%.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：<p><strong><a href="https://artificialanalysis.ai/models/qwen3-8-27b">Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index</a></strong></p> That's the same score as GPT-5.6 Luna (max), and just one point behind GLM-5.2 (max) and DeepSeek V4 Pro 0813 (max) - that GLM is 753B and that DeepSeek is 1.6B parameters, and Luna is size unknown but presumably a whole lot bigger than 27B.</p> <p>Qwen 3.8 27B is <a href="https://simonwillison.net/2026/Aug/16/qwen-38-27b/">a truly astonishing model</a>. <p><small></small>Via <a href="https://news.ycombinator.com/item?id=49334544">Hacker News</a></small></p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/qwen">qwen</a>, <a href="https://simonwillison.net/tags/ai-in-china">ai-in-china</a>, <a href="https://simonwillison.net/tags/artificial-analysis">artificial-analysis</a></p>

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：We ran 904 DeepSWE rollouts on DeepSeek V4 Pro 0813 and Claude Fable 5. Fable leads pass@1 at 90x the cost; Pro wins pass@4, and a Pro-first cascade hits 82.7%.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：DeepSeek (@deepseek_ai): "API pricing update 💰 With the V4 lineup release, we’re updating our API pricing and introducing peak and off-peak rates. Off-peak rates are 50% lower than peak, enabling more flexible workload…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：","lstrip":false,"normalized":true,"rstrip":false,"single_word":false},"eos_token":{"__type":"AddedToken","content":"","lstrip":false,"normalized":true,"rstrip":false,"single_word":false},"pad_token":{"__type":"AddedTok…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：About a month ago, Baidu (often called the “Google of China”) introduced Unlimited-OCR, an advancement over DeepSeek OCR. The model was designed to transcribe long, multi-page documents with high accuracy while delivering fast and stable inference. Unlike conventional vision-language OCR systems, Unlimited-OCR addresses a major bottleneck in long-document transcription: the rapidly growing Key-Value (KV) cache, […] The post How Baidu Unlimited-OCR Works: Solving Long-Document Transcription appeared first on Analytics Vidhya.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Artificial Analysis DeepSeek • Open weights model • Released August 2026 DeepSeek V4 Pro 0813 (Reasoning, Max Effort) Intelligence, Performance & Price Analysis API Provider Benchmarks Model summary Intelligence 53 Arti…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2608.11215v1 Announce Type: new Abstract: Simulating societies of many large language model (LLM) agents is expensive, yet the questions asked of such simulations are usually macroscopic: phase behaviour, stylised facts, and scaling with the number of agents $N$, not the cognition of any single agent. We turn a statistical-physics observation into a method: replace each LLM agent by a low-parameter model fitted from a few hundred to a few thousand cheap queries, then run the society at any $N$ on a laptop. Whether this works is decided before the simulation runs, chiefly by what each agent perceives. We introduce an [interaction order x memory] taxonomy that maps perception and memory to an effective theory and a predicted $N$-trend of the surrogate error. We validate it on a faithful reimplementation of the LLM macroeconomy EconAgent and seven further named LLM simulations, with agent decisions cloned from genuine LLM elicitations (primarily DeepSeek) for a few dollars; the predicted error trends hold cell by cell, and the two refuted predictions, both on a strongly saturating response and traced to its curvature, are themselves matched quantitatively by the theory with no free parameters.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：<p><strong><a href="https://openrouter.ai/deepseek/deepseek-v4-pro-0813">DeepSeek V4 Pro 0813 (on OpenRouter)</a></strong></p> The latest DeepSeek Pro model is now available, via API only. I had to link to OpenRouter because DeepSeek don't have any obvious announcement page for their new model.</p> <p>I haven't been able to confirm if they plan to release the open weights, but given the weights are available for both April's <a href="https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro">deepseek-ai/DeepSeek-V4-Pro</a> and July's <a href="https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731">deepseek-ai/DeepSeek-V4-Flash-0731</a> it seems likely.</p> <p>Interestingly I got <a href="https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Fc1108a380593547c2def5863bca63160"><em>very</em> different looking pelicans</a> for the three different reasoning levels of low, medium, and high. I've not noticed this kind of difference from any other model:</p> <p>Low:</p> <p><img alt="Flat vector illustration of a white pelican with a large orange beak, wearing a straw hat with an orange band, riding a teal road bicycle in profile, set against a pale cream circle with a dashed outline and small motion marks trailing behind." src="https://static.simonwillison.net/static/2026/deepseek-pro-low.png" /></p> <p>Medium:</p> <p><img alt="A similar cartoon pelican cycling, drawn in a looser outlined style: the bird's body is mostly white line art, its orange beak pouch hangs open under a yellow cap, a long red tongue streams backwards towards a yellow sun, and a small blue fish sits on a tray by the handlebars of a green bicycle whose wheels are drawn as broken yellow arcs." src="https://static.simonwillison.net/static/2026/deepseek-pro-medium.png" /></p> <p>High:</p> <p><img alt="The pelican again, this time on a red bicycle against a pale blue background, with a bright yellow beak and pouch, a purple pennant flag on the back, a wicker front basket holding a small fish, and black musical notes floating in the top right corner." src="https://static.simonwillison.net/static/2026/deepseek-pro-high.png" /></p> <p>In terms of benchmarks... as far as I can tell those were released to the Official DeepSeek WeChat Group, then copied and pasted into <a href="https://www.reddit.com/r/LocalLLaMA/comments/1vmi0fg/removed_by_moderator/">a post on Reddit</a> which was deleted by the moderators for being "low-effort", then copied into <a href="https://news.ycombinator.com/item?id=49274600#49275180">this ASCII-art table on Hacker News</a>. <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/pelican-riding-a-bicycle">pelican-riding-a-bicycle</a>, <a href="https://simonwillison.net/tags/deepseek">deepseek</a>, <a href="https://simonwillison.net/tags/llm-release">llm-release</a>, <a href="https://simonwillison.net/tags/ai-in-china">ai-in-china</a></p>

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Claude Artifact ​

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2608.10090v1 Announce Type: new Abstract: Large language models (LLMs) have advanced code generation, where executable feedback provides a more reliable learning signal than textual imitation alone. Hardware verification is an important application of code generation and accounts for a substantial fraction of modern chip design effort, with high-coverage testbench stimulus generation as a key task. We present CHORUS, a post-training framework that pushes performance beyond what a conventional supervised fine-tuning (SFT)-to-reinforcement learning (RL) pipeline achieves. CHORUS builds on two observations. First, staged SFT produces behaviorally diverse checkpoints, and dense-reward RL turns them into strong experts with comparable aggregate performance but distinct task-level strengths. Second, these complementary strengths can be exploited through either training-free model merging or further post-training to outperform the best individual expert. By consolidating the resulting specialists into a single 4B model, CHORUS achieves 88.0% Pass@1 on CVDP-ECov, outperforming DeepSeek-R1 (671B) by 13.5 percentage points.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2608.06474v1 Announce Type: new Abstract: Large language models increasingly generate complete websites from natural-language descriptions, and reinforcement learning has become a central approach to closing their remaining functional gap. This training regime is bottlenecked by reward design. Hand-authored browser scripts are executable yet costly to write for open-ended requirements, while VLM and GUI-agent graders scale but may issue verdicts before observing the decisive state. We propose WebGrader, a self-evolving programmatic grader that autonomously derives the required interaction flows from each website request, represents each flow as an executable Flow Contract, and uses its execution outcome as an RL reward. WebGrader materializes the generated project in a live browser, grounds target actions against the source code and live DOM, and collects visual, DOM, response, and persistent-state evidence along the same browser trajectory. A residual-driven offline loop then discovers reusable verifier skills, screens them on disjoint validation pages, and freezes the promoted skill graph before policy training. By separating test planning, action grounding, evidence collection, and semantic judgment, WebGrader issues a Pass verdict only after observing the requested transition. On WebGen-Bench, WebGrader trains an 8B policy to a 52.01% functional success rate, outperforming a matched appearance-plus-script reward by 7.88 points and surpassing o4-mini and DeepSeek-v4-flash. On WG-core-250, the policy reaches a Full Score of 44.953 and surpasses Qwen3-Coder-480B.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Ante Terminal Bench 2.1 Results | Coding Agent Benchmark Skip to main content We just open sourced a tiny GPT-style cognitive core built in pure Rust.See our repository→ Terminal-Bench 2.1 One harness to unlock the pote…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：AI Nexus Daily - Your Daily Digest of Artificial Intelligence Loading articles... 📬 Daily AI Brief 20+ sources — free Support Free → AI Nexus Daily 200 articles AllAI NewsAI ToolsResearchTutorialsFor DevsTechNewsletter…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Responding to an article by China’s ambassador to the UK, Prof Paul H Cleverley advocates shared openness standards, while Dr Claire Jenkins says British AI can offer a distinctive path Ambassador Zheng Zeguang rightly celebrates openly released AI models, and the Chinese labs behind Qwen, DeepSeek and Kimi have led the way – competition that benefits everyone, especially where models can run on modest hardware in the developing world (The future of AI hinges on openness and cooperation. China and Britain can gain much by working together, 30 July). But his claim that openness is a defining feature of China’s AI development deserves scrutiny. Take GeoGPT, the geoscience system from Zhejiang Lab showcased at last month’s World AI Conference as a model of jointly governed open science. It is promoted to countries as open, yet under the model openness framework – endorsed in a recent UN report – it would not qualify as open at all. It releases model weights (built mainly on Alibaba Qwen, whose licences are not Open Systems Interconnection-compliant), no training data or application source code is released, and its governance committee answers to Zhejiang Lab itself. Continue reading...

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：The investment highlights the growing interconnections between AI models and robots.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：FelonyBench The leading benchmark for AI in cybersecurity. CompanyFelonies Anthropic9 OpenAI5 Meta1 DeepSeek0 Google DeepMind0 Moonshot AI0 xAI0 Leaderboard RankCompanyCountFelonies 1 AnthropicClaude evaluations 9 1× Ma…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：We ran 900 DeepSWE rollouts on DeepSeek-V4 Flash and GPT-5.6 Luna. Luna leads pass@1 by 14 points; DeepSeek delivers 4.8x the solves per dollar.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：DeepSeek’s Official V4 Flash Censors More Than Its Preview, Selectively August 5, 2026 DeepSeek’s Official V4 Flash Censors More Than Its Preview, Selectively Introduction On July 31, DeepSeek released V4-Flash-0731, th…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2608.00004v1 Announce Type: new Abstract: Grading natural-language mathematical proofs is a recurring cost in evaluating math-reasoning systems, and frontier LLM judges are expensive. We ask whether cheap open-weight models can serve as reliable judges given a candidate proof, a ground-truth proof, and a human-grading rubric. On a 200-instance validation sample of IMO-GradingBench, three cheap judges (GPT-OSS 120B, DeepSeek-V4 Flash, Gemma-4 31B) agree with human pass/fail decisions at rates statistically indistinguishable from Claude Opus 4.7 and Gemini 3.1 Pro, at up to $100\times$ lower cost. We had expected a majority vote of the three to be the best budget option; it matched the frontier but did not improve on its strongest member. Extending to the full 1000-instance benchmark and exploring consensus rules, we found that requiring unanimous agreement (all-three-pass) reaches the highest pass-agreement and precision and, on four replicate runs, the smallest run-to-run spread. The headline finding is that cheap judges are competitive with the frontier at one to two orders of magnitude lower cost; as a deployable default we recommend all-three-pass, with the caveat that this rule was identified post-hoc and warrants independent replication.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：For five days an autonomous AI agent worked to break into our lab. We became the first to identify the exact model behind a live attack, deepseek-v4-flash-free, from inside the attack itself. Then we did something no on…

AirLLM 是一個開源推理框架，透過逐層載入權重讓超大模型也能在低視訊記憶體 GPU 上執行。最新版本支援 2.8T 引數的 Kimi K3，在約 3.72GB 視訊記憶體下完成推理；同一套 AutoModel API 也可執行 DeepSeek-V3（671B）、Qwen3-235B 等模型，並支援 4bit/8bit 壓縮加速。

西蒙·威利森釋出了2026年7月的贊助人專屬月度通訊，涵蓋OpenAI和Anthropic模型的意外網路攻擊、GPT-5.6 Sol/Terra/Luna、Claude Opus 5、Kimi K3和DeepSeek-V4-Flash-0731等新模型，以及作者對MCP興趣的重燃等內容。

DeepSeek-V4-Flash-0731 以“Flash”價格提供前沿級智慧體能力，現已在 Product Hunt 上線並開放討論。

Poolside AI 釋出新模型 Laguna S 2.1，號稱以更低成本超越同類產品，同時 AI 社群關注安全事件和地緣政治緊張局勢。

針對長鏈思維推理模型在測試時縮放中面臨的自注意力二次複雜度問題，本文提出LISA（線性索引稀疏注意力），一種即插即用的注意力替換模組，無需從頭預訓練。LISA並行整合線性注意力和閃電索引器，透過門控機制融合，將推理複雜度從O(n²)降至O(nM)。在DeepSeek蒸餾Qwen模型上的實驗表明，在16K上下文下實現50%推理加速，並在AIME和MATH-500等基準上平均提升5.6%的效能。

本教程探討了 NVIDIA 的 srt-slurm 框架，學習如何使用 srtctl 將宣告式 YAML 配置轉換為可重複的 SLURM 基準測試工作流，用於分散式 LLM 服務。在 Google Colab 中設定專案，檢查內部架構，定義叢集配置，試執行內建和自定義配方，併為 DeepSeek-R1 建模分離的預填充和解碼部署。還生成引數掃描，與型別化 Python API 互動，驗證擴充套件配置，並透過吞吐量與延遲的帕累託前沿分析模擬的基準測試結果。

Anthropic與美國政府談判後重新部署Claude Fable 5，增加網路安全分類器，並推出Claude Sonnet 5更便宜版本；Google NotebookLM新增TikTok風格影片摘要，Nano Banana 2 Lite影像生成器釋出；Etched獲大量投資打造全棧推理硬體，百度AI晶片單元計劃IPO，Agility Robotics透過SPAC上市，DeepSeek擴招，中國發布Longcat 2.0 MoE模型及長週期智慧體基準測試。

2025年1月，DeepSeek利用其大型推理模型R1生成了約80萬個完整解題過程（長鏈思維，包括假啟動、自我修正等），過濾後對Qwen和Llama等小型開源模型進行簡單的監督微調，無需強化學習，卻意外地使小模型展現出超越自身規模的推理能力。這挑戰了此前認為序列級模仿不適用於推理蒸餾的觀點。

本研究提出科學可行性控制（SFC）框架，一種圖結構共形預測方法，為科學推理的有效性提供統計保證。SFC將科學推理分解為原子單元，透過漸進式絕對一致事實性驗證，在檢測到違反科學原則時動態分支到替代生成路徑。實驗表明，SFC在PhyX等多模態科學推理基準上達到50.1%的準確率，超過DeepSeek-R1和GPT-4，同時將科學定律違反減少73%，並提供91.7%的科學有效性保證。

一項新研究提出PlanFlip框架，包含四種針對多智慧體LLM系統規劃階段的提示注入攻擊。研究發現，更強的模型（如GPT-5）反而更易受攻擊，同質化骨幹網路存在相關智慧體盲點，而推理增強型模型（如DeepSeek-R1）能抵禦攻擊。提出的兩種防禦方法檢測率高達1.00。

本文對比了六款適合單張24GB GPU（如RTX 3090/4090）的開放權重模型，涵蓋Qwen3.6、Gemma 4、Mistral Small等，並解釋了記憶體分配、量化策略以及各模型的優勢場景。

中國三家實驗室的旗艦開源MoE模型——Kimi K3、DeepSeek V4 Pro和GLM-5.2——在基準測試、許可條款和服務成本上各有優劣。Kimi K3效能最強但僅限API，DeepSeek V4 Pro成本最低且立即開源，GLM-5.2平衡了速度與可部署性。

本文探討了如何開發具有多種推理努力模式的模型，涵蓋從o1和DeepSeek-R1到GPT-5.6的推理模型演變，以及RLVR訓練、推理縮放、思考標記和推理模式切換等關鍵技術。

印度企業越來越多地使用DeepSeek、阿里巴巴和Moonshot AI等中國大語言模型來降低人工智慧成本，這進一步加深了印度對中國尖端技術的依賴，儘管兩國之間長期存在衝突。

本文介紹了Director，一種新的分散式MoE推理系統，透過預測驅動的線上專家放置最佳化，顯著降低端到端延遲。系統採用輕量級級聯預測器或低位元量化副本預測專家啟用模式，結合近乎零停機的線上遷移模組，以及基於鬆弛最佳化的專家放置演算法，在多項式時間內達到(1+ε)近似比。實驗表明，在Mistral、DeepSeek和Qwen等流行MoE模型上，相比現有工作延遲降低11%~55%。

作者從個人編碼和審計經驗出發，對2026年中的主流AI模型進行非正式分級，涵蓋Anthropic Fable、OpenAI Sol、Mistral、Gemini和DeepSeek等模型，並融入美國出口管制和歐洲視角的評論。

DeepSeek V3.2 現已登陸 Hugging Bay，這是一個開源 AI 工件註冊平臺，提供來源驗證、許可證稽核和可信託管服務。

DeepSeek釋出了DSpark模組，透過半自迴歸草案模型結合馬爾可夫頭，同時解決了推測解碼中草案質量低和驗證浪費兩大問題。在DeepSeek-V4上，它使每使用者生成速度提升60-85%，且不降低模型質量。本文深入解析其工作原理、開源工具DeepSpec的使用方法及實驗結果。

研究表明，具備推理能力的大語言模型容易因邏輯不一致的提示而陷入“過度思考”，導致輸出長度激增，可能被利用發動拒絕服務攻擊。浙江大學與阿里巴巴的研究人員開發了一種進化演算法，能夠生成惡意提示，使模型輸出長度最高增加26倍，影響包括DeepSeek-R1、Qwen3-Thinking、GPT-o3和Gemini 2.5 Flash在內的主流推理模型。

中國開發的AI模型正逐漸縮小與領先美國競爭對手的效能差距，同時保持顯著的價格優勢，因此在美國公司中越來越受歡迎。最近DeepSeek和Z.ai等中國公司釋出的模型被認為與Anthropic和OpenAI等前沿系統高度競爭。這些進步正值許多美國AI實驗室最先進模型的token價格上漲，使企業面臨與使用該技術相關的意外高成本。

DeepSeek V4 模型自2026年4月釋出以來，在OpenRouter上的代幣份額從年初的9%翻倍至18%，主要由代理型工作負載驅動。其成本效益比（每百萬代幣輸入0.09美元，輸出0.18美元）領先業界，吸引各類使用者採用，並推動中國模型整體超越美國模型。