{"rewrite":{"id":"r_533435eeee7a6769512501ee","clusterId":"c_734c873702ca8bc34f8ee4f7","slug":"deepseek-quietly-releases-v4-pro-0813-model","model":"deepseek-v4-flash:free","headline":"DeepSeek Quietly Releases V4 Pro 0813 Model","summary":"Chinese AI startup DeepSeek has released its latest AI model, DeepSeek V4 Pro 0813, without a formal announcement. The release was first noticed on social media and message boards, where users and tech commentators highlighted the model's performance and cost efficiency. According to the open-source AI coding agent Cline, the model scores 15.8 percent higher on Terminal Bench, a benchmark that evaluates how accurately AI can execute real system operations and development tasks on a Linux terminal, compared to the April preview model. Cline also noted that DeepSeek V4 Pro 0813 runs at roughly one fifty-seventh the cost of Claude Fable 5 while delivering comparable performance. The model has 1.6 trillion parameters, 49 billion active parameters, and a 1 million token context window. Tech writer @ChrisGPT reported benchmark improvements across Terminal Bench 2.1, CyberGym, and DeepSWE, with scores rising from 72.1 to 87.9 percent, 52.7 to 83.3 percent, and 12.8 to 62.7 percent respectively. Pricing is set at 0.435 dollars per million input tokens and 0.87 dollars per million output tokens, with a discounted cache-hit input rate of 0.003625 dollars.","whyItMatters":"The quiet release, paired with a steep price cut versus rival models, signals DeepSeek is pushing cost-performance as its main competitive lever in the AI model market.","webCardHtml":"\u003cp\u003e\u003c/p\u003e\u003cul\u003e\n  \u003cli\u003eDeepSeek V4 Pro 0813 appears in the official DeepSeek API documentation under the name \u0026#34;DeepSeek-V4-Pro-0813,\u0026#34; confirming the release.\u003c/li\u003e\n  \u003cli\u003eCline, the open-source AI coding agent that first flagged the model, called it \u0026#34;the best price-to-performance model currently available on the market.\u0026#34;\u003c/li\u003e\n  \u003cli\u003e@ChrisGPT, a writer for The Information and TechCrunch, posted the benchmark and pricing figures that circulated on social media.\u003c/li\u003e\n  \u003cli\u003eGIGAZINE ran its own test asking the model to draw \u0026#34;a pelican riding a bicycle\u0026#34; in SVG format; the results were published alongside the release coverage.\u003c/li\u003e\n  \u003cli\u003eThe release follows DeepSeek\u0026#39;s pattern of quiet launches: the company previously released \u0026#34;DeepSeek-V4-Flash-0731,\u0026#34; an open model priced below GPT-5.6 Luna with comparable performance, without a formal announcement.\u003c/li\u003e\n  \u003cli\u003eDeepSeek has also open-sourced \u0026#34;DeepSeek Harness\u0026#34; v0.1 under the MIT license, a harness for AI agents built on the meta-framework Cordis, where models, tools, skills, sessions, sandboxes, and UI are all implemented as plugins.\u003c/li\u003e\n  \u003cli\u003eDeepSeek Harness records all model perceptions, from prompts to reasoning and tool calls, in an append-only session log, and offers four execution modes: Standard, Code, Minimal, and Creator.\u003c/li\u003e\n  \u003cli\u003eThe US government\u0026#39;s AI risk management agency CAISI previously reported that DeepSeek V4 Pro trails major American AI models by about 8 months but is currently the most powerful Chinese-made AI model.\u003c/li\u003e\n\u003c/ul\u003e\u003cp\u003e\u003c/p\u003e","blueskyPost":"DeepSeek V4 Pro 0813 ships without fanfare, yet Cline's benchmark shows a 15.8% jump on Terminal Bench. The quiet rollout suggests DeepSeek prioritizes efficiency over hype, aiming for cost-conscious developers.","twitterPost":"DeepSeek V4 Pro 0813 launches silently, but Cline reports a 15.8% Terminal Bench gain at 1/57th of Claude Fable 5's cost.","threadsPost":"DeepSeek V4 Pro 0813 appeared with no announcement, yet Cline's Terminal Bench scores jumped 15.8%. At 1/57th the cost of Claude Fable 5, DeepSeek is betting on efficiency over spectacle, targeting developers who watch their API bills.","newsletterBlurb":"DeepSeek quietly released its latest AI model, V4 Pro 0813, with big benchmark jumps and a price that undercuts rivals. The model is already listed in the official API docs, and early commentary highlights its cost-performance ratio.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260813-deepseek-v4-pro-0813/\",\"title\":\"中華AIスタートアップのDeepSeekが「DeepSeek V4 Pro 0813」をひっそりとリリース\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":11927,"outputTokens":1403,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1786697630,"createdAt":"2026-08-14T08:50:11.000Z","publishedAt":"2026-08-14T08:51:45.000Z","updatedAt":"2026-08-14T08:50:11.000Z"},"cluster":{"id":"c_734c873702ca8bc34f8ee4f7","canonicalTitle":"中華AIスタートアップのDeepSeekが「DeepSeek V4 Pro 0813」をひっそりとリリース","representativeArticleId":"a_f0183d57ad29cb9e4f3d38b1","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":false,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"DeepSeek V4 Pro 0813\"],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-08-13T06:25:00.000Z","lastSeenAt":"2026-08-14T07:31:00.000Z","updatedAt":"2026-08-14T08:51:45.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260814-deepseek-harness-v0-1/","title":"DeepSeekがAIエージェントのハーネス「DeepSeek Harness」v0.1を公開"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["DeepSeek V4 Pro 0813"],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["DeepSeek V4 Pro 0813 has 1.6 trillion parameters, 49 billion active parameters, and a 1 million token context window.","Pricing is 0.435 dollars per million input tokens and 0.87 dollars per million output tokens, with a cache-hit input rate of 0.003625 dollars.","Terminal Bench 2.1 scores improved from 72.1 percent to 87.9 percent, CyberGym scores from 52.7 percent to 83.3 percent, and DeepSWE scores from 12.8 percent to 62.7 percent.","Cline reported the model runs at about one fifty-seventh the cost of Claude Fable 5 with comparable performance.","The model is listed in the official DeepSeek API documentation as DeepSeek-V4-Pro-0813."]}