DeepSeek V4 "full-power version" has been exposed and is set to be released as early as tomorrow.

The entire online community has been waiting for nearly three months!

The official release of DeepSeek V4 is expected to launch as early as tomorrow, and no later than within the next few days.

At present, a portion of users have already been granted early access to the gray-scale testing of DeepSeek V4 (GA).

There are two versions in total: DeepSeek V4 Flash, and DeepSeek V4 Pro.

The "Full-Power Version" of DeepSeek V4

Is Finally Arriving

What everyone is most concerned about must be: how to check if you have been included in the gray-scale test group?

Tech blogger AiBattle shared a popular folk "verification trick": check the first-person perspective in the Chain of Thought (CoT).

If the model's thinking process starts with "I'm" or "I'll" instead of the old version's opening "Let me".

Congratulations, you are very likely already using the V4 GA version!

At a time when Fable 5 and GPT-5.6 Sol are locked in continuous competition, the entire online community has extremely high expectations for this major breakthrough in open-source AI.

After testing, developer Pankaj Kumar first gave a fair summary of V4's performance:

Its overall performance is close to the Opus 4.8 level, and its coding capability directly competes with GPT-5.6 Sol;

Agent capabilities have been greatly enhanced, and 3D and SVG generation performance has significantly improved;

For the exact same task, V4 requires more iteration rounds than Fable 5.

He stated that from the current market positioning, V4 is very likely not to outperform the newly released Kimi K3, but its price will be significantly lower.

If this level of performance is truly paired with this price point, it could very well mark another iconic DeepSeek moment.

First Round of Testing Results Are Out

Nowadays, demo clips from the first round of testing for DeepSeek V4 (GA) have begun to circulate publicly.

Public feedback on the model is mixed:

Some users believe it can already match the performance of Claude 5, while some developers point out that the Pro version does not offer a huge performance leap over the Flash version.

The following is a 3D simulated shooting game generated by V4 Pro. Its core gameplay involves operating a siege crossbow vehicle to complete target practice, with all basic UI features fully implemented.

The official V4 release has created an HTML-based hybrid game combining elements of *Minecraft* and *No Man's Sky*, which offers a fairly high level of playability.

The classic game *Cut the Rope* shown below was also generated in a single run by V4.

The following are additional demos generated by V4: an SVG test for Xbox controllers, and another game generation work.

Peak-Valley Pricing Introduced for the First Time

Still an Incredible Value

The biggest variable in creating another "DeepSeek moment" lies in its pricing strategy.

All leaks point to the same conclusion: performance may not be the absolute top-tier, but the price will be significantly lower than competitors.

At the end of last month, DeepSeek sent an email to all API users, announcing that the official V4 release would be launched in mid-July.

API pricing will be adjusted simultaneously with the GA release, introducing the new "peak-valley pricing" mechanism.

• deepseek-v4-pro: $0.87 per million output tokens during off-peak hours, $1.74 during peak hours; $0.435 per million input tokens for cache misses during off-peak hours.
• deepseek-v4-flash offers even more aggressive pricing: $0.28 per million output tokens, $0.56 during peak hours, and only $0.0028 per million input tokens for cache hits.

In other words, the legendary "price killer" that never raised prices has installed a metering system for its computing power for the first time.

This change has little impact on individual users, but for teams that run Agent tasks nonstop during working hours, this is a cost that needs to be recalculated carefully.

Fortunately, the price for cache hits remains extremely low. By scheduling batch tasks, benchmark testing, and data generation outside peak hours, costs can still be kept very low.

Even so, compared to Fable 5's $50 per million output tokens, V4 still offers the most outstanding cost performance on the market.

After all, its performance has already reached the Opus level. When the V4 preview version was released, official self-test results on SWE-bench Verified showed that —

DeepSeek-V4-Pro-Max was only 0.2 percentage points behind Claude Opus 4.6 Max, while its price was only one-seventh of its competitor's.

However, Opus still maintains a leading position in long-context retrieval, professional software engineering, and some knowledge reasoning benchmarks.

As a side note, the two old model names deepseek-chat and deepseek-reasoner will be officially discontinued on July 24.

In any case, the long-awaited "shoe" that the entire online community has been waiting for three months is finally about to drop.

Based on the currently leaked information, V4 is very unlikely to be the model that ranks first in every single performance metric.

With Fable 5 and GPT-5.6 Sol ahead of it, and Kimi K3 as a close competitor, it will be very difficult for V4 to disrupt the market purely by raw performance.

But DeepSeek has never relied on that strategy.

The real highlight is the proven, repeatedly successful path: delivering Opus-level capabilities at just one-seventh the cost.

After all, in a market dominated by tech giants that charge tens of dollars for millions of tokens, even with the new "peak-valley pricing meter" installed, V4 is still the formidable "price killer" that dominates the competition.

References:

https://x.com/pankajkumar_dev/status/2078536231026372846

This article is from the WeChat public account "AI Era", author: ASI Revelation, editor: Taozi, published with authorization from 36Kr.