{"slug": "260812-deepseek-harness-just-launched-its-official-account-v4-pro-final-version", "title": "[260812] DeepSeek Harness Just Launched its Official Account, V4 Pro Final Version Coming Soon", "summary": "DeepSeek Harness, the Agent toolchain project officially led by DeepSeek, launched its official WeChat account on August 11, signaling the imminent release of the tool and the final version of DeepSeek-V4-Pro. The official team also announced a significant API price increase and peak-valley pricing to manage computing power pressure and high token consumption costs, advising developers to optimize cache hit rates and schedule non-urgent tasks during off-peak hours.", "body_md": "# DeepSeek Harness Just Launched its Official Account, V4 Pro Final Version Coming Soon\n\nTLDR\n\n- DeepSeek Harness official WeChat account launched on August 11, signaling the imminent release of this Agent toolchain, which directly competes with Claude Code. It is highly likely to be bundled with the official V4 Pro release. Meanwhile, to address the computing power pressure and high token consumption costs brought by the explosion of the Agent ecosystem, the official team has announced a significant API price increase and peak-valley pricing. Developers are advised to promptly calculate costs, optimize cache hit rates, and schedule non-urgent tasks during off-peak hours before the price hike, in anticipation of the upcoming product launch and price adjustments.\n\nToday, the DeepSeek Harness team quietly launched its official WeChat account!\n\nSome might still be unfamiliar with the name—**DeepSeek Harness** is an Agent toolchain project officially led by DeepSeek. **You can think of it as DeepSeek's version of Claude Code or Codex.**\n\nThe team started building this project in May of this year, and just began recruiting for closed beta testing on August 1st. Only 10 days later, the official WeChat account was launched as a communication channel. This indicates that **Harness is not far from its official release, and it is highly likely to be deeply tied to another major event—the full rollout of the official V4 Pro version.**\n\n## Connecting the Recent Information \n\nLike a snake in the grass, the official team has actually been laying the groundwork. Connecting a few recent timestamps, you'll find a very tight combination of moves:\n\n**July 31st**: V4-Flash official API launched for public beta. The official announcement included a subtle hint: \"DeepSeek-V4-Pro official version will be released as soon as possible.\"**August 1st**: DeepSeek Harness began quietly recruiting for closed beta testing.** August 6th**: A notification suddenly popped up in the developer backend: plans to collectively increase API service pricing soon,** with a significant expected increase**.** Today (August 11th)**: DeepSeek Harness launched its WeChat official account.\n\n## The Price Increase Announcement That Made Developers Anxious \n\nThe August 6th announcement was concise, containing only one sentence: API pricing will be generally increased soon, with a significant increase expected, and it is recommended to arrange usage reasonably.\n\n**It didn't state how much it would increase, when it would take effect, or which specific model it would affect.** This \"three-noes\" notification made people even more uneasy.\n\nLet's review the current price list (RMB/million Tokens):\n\n| Model | Input (Cache Hit) | Input (Cache Miss) | Output |\n|---|---|---|---|\n| V4-Flash | 0.02 Yuan | 1 Yuan | 2 Yuan |\n| V4-Pro | 0.025 Yuan | 3 Yuan | 6 Yuan |\n\nNote that even with the already decided \"peak period doubling\" plan, the official team still emphasized \"significant price increase.\"\n\n## Why a Price Increase at This Time? \n\n**First, computing power levels are indeed in short supply.** According to data from overseas developer platforms, the daily token processing volume of the V4 Flash model once soared to **8 trillion**—this figure even exceeded the average daily total throughput of OpenRouter, an aggregation platform hosting over 400 models. Even with official limits of 2500 (Flash) and 500 (Pro) concurrent requests, the massive number of requests couldn't be suppressed. Peak-valley pricing is for diversion, and overall price increase is for interception; essentially, both are using price levers to maintain concurrency.\n\n**Second, computing power costs are forcing \"price killers\" to stop.** Looking at the domestic large model market, from Zhipu and Kimi to major cloud providers, prices have been adjusted this year. The shortage of high-end GPUs, high hardware and electricity costs, and the strategy of burning money for scale have hit their ceiling.\n\nLiang Wenfeng once mentioned: when user demand is \"almost inelastic\" to price, a moderate price increase is an inevitable choice to ensure service stability.\n\n**It's not because they want to make more money, but because too many people are using it; without a price increase, everyone would crash together.**\n\n## What Does Harness Have to Do With the Price Increase? \n\nConnecting the price increase and Harness makes the logic complete.\n\nHarness's core capability is to grant models the authority to control browsers, read and write local files, and execute Shell commands. As developers know, each of these automated operations by agents burns tokens: Tool calls themselves have context overhead, and the vast results returned by tools (web DOM structure, lengthy command output logs) must be crammed into the Context Window for the model's next inference. **A complete Agent task often consumes several or even dozens of times more tokens than pure text conversations.**\n\nIf the core killer feature of the V4 Pro official version is Agent capabilities (stronger multi-step reasoning, more precise tool calls), then the overall API usage will experience a true explosion. **The current price increase is precisely to build a dam in advance for the upcoming \"token flood\" of the Agent ecosystem.**\n\nConversely, the Harness team entering the arena now also indirectly confirms that the underlying Agent capabilities of V4 Pro have been refined and are ready to be tested by developers' real business needs.\n\n## What Can Be Done Before the Price Increase? \n\nIf you are using DeepSeek V4 API to run agents, don't wait at this stage; do these three things first:\n\n**1. Calculate your costs clearly; don't regret it after the increase.** Monthly fees vary greatly depending on usage and model combinations. For a project consuming an average of 10 million output tokens daily, the current monthly fee is approximately ¥600 for Flash and ¥1,800 for Pro. If the price doubles, the difference jumps directly to ¥1,200 vs ¥3,600. Decide whether to switch from Pro to Flash or mix multiple models after calculating the costs.\n\n**2. Obsess over cache hit rate.** The price difference between V4 Pro cache hit (¥0.025/M) and cache miss (¥3/M) is **120 times**. Do not resend the same system prompt and tool definition every time. Refactor your Prompt structure to enable caching, and costs will directly drop by two orders of magnitude.\n\n**3. Schedule batch automated tasks during \"off-peak hours.\"** The peak-valley price difference is one-fold; move non-urgent batch processing, data cleaning, and benchmark runs to evenings or weekends, and costs will be directly halved.\n\n## Harness + V4 Pro: What's Next? \n\nIn the next few days, pay close attention to these two indicators:\n\n**Mid-August actions**: Multiple media outlets previously reported that the V4 official version might be delayed until mid-August, most likely to coincide with the closed beta testing of the DeepSeek Harness toolchain. Now that the official account has been launched, closed beta slots could be announced at any time.**Specific price increase implementation**: Once the final pricing is announced, the entire domestic API's price-performance benchmark will be redefined.\n\nI will follow up on further DeepSeek Harness beta developments as soon as possible. After the V4 Pro official version is released, I will also publish an in-depth comparison of its Agent capabilities.\n\n**Follow me to not miss out on first-hand hardcore information for large model developers.**\n\nPlease indicate the source and link of this article when reprinting.\n\nHelp us make these docs great!\n\nAll X-CMD docs are generated from command help and multiple data sources. See something that's wrong or unclear? Feel free to let us know through any of these ways~", "url": "https://wpnews.pro/news/260812-deepseek-harness-just-launched-its-official-account-v4-pro-final-version", "canonical_source": "https://x-cmd.com/blog/260812/", "published_at": "2026-08-12 00:00:00+00:00", "updated_at": "2026-08-13 04:05:02.125965+00:00", "lang": "en", "topics": ["ai-agents", "ai-products", "ai-infrastructure", "ai-policy"], "entities": ["DeepSeek Harness", "DeepSeek", "Claude Code", "Codex", "V4-Flash", "V4-Pro", "OpenRouter", "Liang Wenfeng"], "alternates": {"html": "https://wpnews.pro/news/260812-deepseek-harness-just-launched-its-official-account-v4-pro-final-version", "markdown": "https://wpnews.pro/news/260812-deepseek-harness-just-launched-its-official-account-v4-pro-final-version.md", "text": "https://wpnews.pro/news/260812-deepseek-harness-just-launched-its-official-account-v4-pro-final-version.txt", "jsonld": "https://wpnews.pro/news/260812-deepseek-harness-just-launched-its-official-account-v4-pro-final-version.jsonld"}}