DeepSeek V4 Pro (0813) Is Live on Maliang Writer: 5 Key Points

In this article 7 sections

The DeepSeek V4 Pro official release (DeepSeek-V4-Pro-0813) is live. From late at night on August 12, 2026, Beijing time, into the 13th, the model version number on the official pricing page switched to 0813; the call name is unchanged, still deepseek-v4-pro, and the new weights take effect automatically. Maliang Writer connected to this model long ago, which means you are already on the official release, and there is nothing to change.

When we wrote about the DeepSeek V4 Flash official release, we mentioned that “the V4-Pro official release will be out as soon as possible” was the official wording; now it has been delivered. Most coverage across the web is still about coding and agents; this article covers only what web-novel writers care about: the data, the price, and whether you need to touch your settings.

The Conclusion First: 5 Things

1. The official release is really here. Version number DeepSeek-V4-Pro-0813, with the official pricing page updated; but the full announcement and official-release benchmarks had not yet been issued as of publication, and we will update this article once the official side publishes them.
2. It takes effect with zero action. The model name is unchanged and the weights switch automatically: so it is for API users, and so it is for Maliang Writer users.
3. The official list price hasn’t changed for now: input $0.435, output $0.87 (per million tokens; cache-hit input as low as $0.003625; all official API pricing). But the official warning of a significant price increase to come still looms.
4. The writing-oriented data is worth noting separately: win rate against Gemini-3.1-Pro in Chinese writing 62.7%, quality win rate in creative writing 77.5%, 1M long-text retrieval (MRCR) 83.5, from the V4 technical report, with the basis explained below.
5. The division of labor on the site is unchanged: Flash continues as the daily-update main force, and Pro continues to sit in the final consistency review and the stronger tier for hard chapters.

From Preview to 0813: Timeline

An honest note: as of this article’s publication (August 13, 2026), DeepSeek’s official update log still stops at the July 31 entry, and the official announcement of 0813 and its accompanying benchmarks have not yet been issued. The hardest evidence right now is the version number on the pricing page, and Vercel gateway’s same-basis notice that “the new weights are in effect by default”. After the official side publishes its announcement, this article will get a second update.

Specs and Official Pricing: All in One Table

First put the two V4 siblings side by side. The data comes from the official pricing page (verified August 13, 2026), and all prices are official API pricing:

Two footnotes that concern your wallet: first, the pricing page footnote states plainly that “overall prices will be significantly raised soon”, with the specific price list and effective date to be announced officially; second, per the HuggingFace community analysis of the Flash official release, 2× peak-hour billing (Beijing time 09:00–12:00 and 14:00–18:00) has been announced but is not yet in effect. To sort out your usage structure before the price change, you can first read Smart Context Assembly to Save Tokens.

Writing-Oriented Data: This Time Not Just Coding Stories

The V4 technical report leaves writers a few numbers they can remember directly: in Chinese writing against Gemini-3.1-Pro, a win rate of 62.7% (the opponent at 34.1%), and in creative writing a “writing quality” win rate of 77.5% (36Kr’s analysis of the technical report, Tencent Cloud’s breakdown of the technical report).

A reminder on the basis: these numbers come from the V4 technical report (published during the preview period, some on a high-configuration basis, summarized in the technical report notes), and officially there are no dedicated benchmarks yet for the 0813 official release. Going by the Flash official release’s practice of “structure unchanged, only post-training redone”, the Pro official release most likely follows the same route, but that is speculation, and the exact changes should wait for the official release note.

Translated into web-novel terms:

• Chinese language feel and literary expression are its home turf, not a side ability. In community tests, its period-style dialogue and grasp of xianxia terminology have also consistently been rated high (AIBotGo’s four-tool review).
• A 1M context plus a long-text retrieval score of 83.5 hits the most painful problem of long novels, “continuity errors”: foreshadowing planted in the first 800,000 characters is something it can find again.
• Don’t dodge the weak points either: the technical report’s self-assessment says factual knowledge is weak (SimpleQA-Verified 57.9, vs 75.6 for Gemini-3.1-Pro), so verify plot beats that involve real-world knowledge yourself; the “expository tone” AI flavor still exists in raw first drafts, and before submitting we suggest a manual polish or the submission compliance self-check.

What This Means for Maliang Writer Users

You don’t need to do anything. The platform calls the model name deepseek-v4-pro, and after the upstream weights switch it takes effect automatically; there is no need to touch your model preferences or usage habits.

What’s worth knowing is where it sits on the site:

• In the final review stage of the whole-book consistency check, the default model is V4 Pro. For a job like continuity-error detection, which has to dig through old accounts across hundreds of thousands of characters, the 1M context and long-text retrieval score are made for exactly this; for usage see the Whole-Book Consistency Check Guide.
• In everyday scenarios such as chapter body text and blueprint generation, V4 Flash is the default, and Pro is the stronger tier for hard chapters and high-precision scenarios. The background of this division of labor is spelled out in the Flash official release article.
• Platform-side usage also supports this layering: in early August we published that V4-Pro’s settled tokens in July were about +1.8× June’s, so the stronger tier is getting busy, not gathering dust.

Should you switch your daily-update main model to Pro too? We suggest the same as a month ago: don’t. The pairing of Flash for daily updates and Pro for hard chapters is still the best value. For the systematic selection logic, see the AI Writing Model Selection Guide, for all models available on the site see the AI Models feature page, and if you’re still torn between “using DeepSeek’s official site directly” and “using a writing platform”, the Maliang vs DeepSeek comparison page breaks down that math.

By the way: within the same 24 hours, xAI also released Grok 4.6. Should novel writers care? Our judgment is in this analysis.

FAQ

Q1: When was the DeepSeek V4 Pro official release published?

From late at night on August 12 into August 13, 2026, Beijing time. The model version number on the official pricing page has been updated to DeepSeek-V4-Pro-0813, and gateways such as Vercel likewise announced that the new weights are in effect by default. The full official announcement had not yet been issued as of August 13, and this article will be updated after it is.

Q2: How does the official release differ from the preview?

Officially no notes have been published yet. Going by the practice of the V4-Flash official release (0731), where model structure and size stay the same and only post-training is redone, the Pro official release most likely takes the same route, but that is reasonable speculation, and the exact changes are subject to the official release note.

Q3: Has the price changed? Will it go up?

For now the official API pricing is unchanged: input $0.435, output $0.87 per million tokens, cache-hit input $0.003625. But the official side has announced that “overall prices will be significantly raised soon”, and there is also 2× peak-hour billing that has been announced but is not in effect; the specific price list and effective dates have not been published.

Q4: Do I need to reconfigure anything in Maliang Writer?

No. The model name the platform calls is unchanged, the new weights take effect automatically, and neither your model preferences nor your usage need to change. Books already using V4 Pro-related features (such as the final review of the consistency check) are automatically upgraded to the official release.

Q5: Flash or Pro for writing novels?

Choose Flash for daily body text, dialogue chapters and progress along the outline; use Pro for wrapping up several threads, whole-book consistency checks, and hard chapters that repeated rewrites can’t get right. In one sentence: Flash keeps updates alive, Pro saves the day. For a detailed comparison table, see the Flash official release article.

Closing

From the April preview to the August official release, this generation of DeepSeek V4 has filled in both foundations needed for writing long Chinese novels: first-tier Chinese writing data, and 1M-level memory. For Maliang Writer users, the official release landing is a quiet good thing: with nothing to do, the foundation has become one notch sturdier.

The price-increase notice is the next point to watch, and on the day the official side publishes the price list, we will follow up with an analysis.

Related reading:

• DeepSeek V4 Flash Goes Official: Why Web-Novel Writers Should Use It
• Grok 4.6 Released: Do Novel Writers Need to Care?
• Which Model for AI Novel Writing in 2026? Tested Rankings
• Whole-Book Consistency Check Guide
• Smart Context Assembly to Save Tokens