The short answer

DeepSeek hasn't confirmed these are V4. What we have is a large, unusually consistent wave of suspected gray-test output: 121 Bilibili videos from 53 creators between Jul 7 and Jul 19, 2026, plus 40 shared OpenCode conversations and a handful of downloadable projects. That's enough to spot patterns — not enough to certify a model ID or run a controlled benchmark.

Read the evidence that way and the picture is clear: V4 looks slightly behind Kimi K3, GPT-5.6 and Fable 5.0 in their strongest pockets, but close enough that, for most daily coding, a cost-conscious developer could default to V4 and keep one light premium subscription for review, nasty bugs, and the last 10% of long multi-turn work.

Practical verdict: the suspected build doesn't write flawless code — the shift is that it can turn one line of intent into a running app, then fix a useful share of its own bugs. That's the threshold between "autocomplete" and "implementation agent".

For the screenshot gallery and chat links from the earlier wave, see our DeepSeek V4 gray-release evidence roundup.

Full source catalog Key evidence

The raw evidence behind every claim in this article — mirrored from the open YunhaoFu/dsv4ga-news-gather repository. Expand to browse by creator, date or keyword. Titles link straight to Bilibili.

Click to expand all 136 records 121 test videos + 15 related posts

Tip: the title is the link to Bilibili. = shared chat session, orange = project download. Hover a creator to see the full name.

#DateCreatorTitle / evidence
1 07-19 不爱亲 疑似deepseekV4灰度正式版,一句话生成的mc光影包
2 07-19 一只AI风向标 DeepSeek V4正式版泄露、Opus 5下周发布或由Karpathy训练、京东发布JoyAI-Talker与Video-Edit | 7月19日 AI日报
3 07-19 猫伊cat 【Deepseek】D老师劝用户不要人机恋之后心态崩了然后搓了个这个
4 07-19 Ai产品小丁 DeepSeek V4 正式版要来了?
5 07-19 进化中的阿陈 Kimi K3.1、DeepSeek V4、Grok 4.6接连放出新爆料,Fable 5 却还在限制 50% 用量 !
6 07-19 kdzzzds ds灰度-(不搜素材 +SVG一轮出) 高质量gta5 2.0
7 07-19 帅气的丝庐海椮潔 DeepseekV4一句话生成的类cs游戏
8 07-19 月隐隐约 Deepseek v4正式版文字生成游戏视频合集
9 07-19 星光与雪 DeepSeek v4正式版一句话设计生成减速器
10 07-19 小鲤玩游戏_ DeepSeek V4 正式版即将发布,有哪些值得关注的亮点?
11 07-19 Augenstern_-__- dsv4灰度正式版-再次进化
12 07-19 咖喱鱼 V4一拳崩死FABLE,中国模型无限追杀A÷和GPT
13 07-19 小小小名不是小明 DeepSeek V4 疑似正式版 两句话生成气液CFD网页(较为不真实)
14 07-19 小小小名不是小明 DeepSeek V4 正式版,帮我生成一首电子音乐【基于strudel.cc代码生成的音乐三首】
15 07-19 冬眠の松鼠_ DeepSeek V4正式版 炉石传说
16 07-19 北宸Polariss 【deepseek v4正式版】土豆兄弟
17 07-19 夜之叶归一 deepseek v4 pro 正式版建模测试
18 07-19 Delight-linger 工程实战?【dsv4对战k3】3D物理骨骼动画,对比让人核爆
19 07-19 北宸Polariss 【deepseek v4正式版】水果忍者
20 07-19 UPLUZ 【疑似DeepSeek正式版】灾害模拟,比前几天看到的又进化了
21 07-19 戏场上的老将军 deepseekv4正式版一段提示词生成黑洞
22 07-19 bili_3493108501187431 deepseekv4正式版战雷最终测试版
23 07-18 -_一-_一- 用deepseek正式版的竞争对手gpt5.6Sol花费很多句话实现的大战略地图
24 07-18 PotnQ DSV4灰测一款游戏迭代了几百次有多恐怖?
25 07-18 戏场上的老将军 deepseekv4正式版 超视距8192格MC
26 07-18 UPLUZ 【疑似DeepSeek正式版】谁才是模型之王啊?(战术后仰)
27 07-18 一只AI风向标 DeepSeek V4正式版灰度测试、马斯克Grok 4.6下周开测、上海AI实验室Intern-S2击败Opus4.8 | 7月18日 AI日报
28 07-18 UPLUZ 【疑似DeepSeek正式版】我和D指导重新设计了像素鸟
29 07-18 cacl2_lhg DeepSeek V4 正式版灰度测试制作的我的世界版马车(含多种不同赛道)
30 07-18 宇智波华人 用deepseek做的小游戏,三国斗飞机!!!
31 07-18 梦回0 dsv4正式版还原《星空Starfield》1000颗星球可无缝起降,陶德你的梦想deepseek帮你实现了!
32 07-18 爱摸鱼的月萌 【中配】Anthropic 发展遇冷|Kimi K3.1、Grok 4.6、DeepSeek v4 正式版、美国政府 Gold Eagle 模型,还有机器人综合
33 07-18 赤麟方阵2号 deepseek灰度测试,地球防卫军试玩
34 07-18 bili_3493108501187431 deepseekv4正式版生成的战争雷霆优化版
35 07-18 硅谷神奇 Anthropic的变故……Kimi K3.1、Grok 4.6、DeepSeek v4正式版、美国政府“金鹰”计划,以及机器人综合格斗!
36 07-18 山头小妖 用deepseek,一天开发游戏的进度(新游戏开发中:末日堡垒对抗)
37 07-18 Aki_2519 [疑似Deepseek v4正式版]类Teardown游戏生成
38 07-18 UPLUZ 【疑似DeepSeek正式版】这才是真正的音乐.jpg②
39 07-18 紫巅 (灰测)DeepSeek正式版一次生成重生细胞
40 07-18 小小小名不是小明 双叉臂悬挂测试后续 没有人可以诋毁V4正式版
41 07-18 烟神殿殿主 AI大洗牌!Anthropic跌下神坛,Kimi K3.1、Grok 4.6、DeepSeek v4大混战,还有美政府金鹰计划与机器人MMA
42 07-18 TwoShenMengxi deepseek正式版无人深空X我的世界(目前已坠机)
43 07-18 Delight-linger 【dsv4正式版】用Godot与C#一句话复刻[网吧模拟器]游戏
44 07-18 UPLUZ 【疑似DeepSeek正式版】震惊瘫坐!节奏光剑都被他搞出来了!
45 07-18 杠精MrS deepseekv4pro max疑似灰度正式版,一句话生成合金弹头,引擎也是自己写的无敌了
46 07-18 炽热的钢铁 疑似灰度到了deepseek-v4-pro正式版!只用一句话让大肥鱼手搓了5首电音!
47 07-18 夜之叶归一 deepseek V4 正式版 测试
48 07-18 kdzzzds ds灰度 超视距8192格MC 1轮出+2轮修bug优化
49 07-18 雨下冰梧桐 请让我真诚的说DEEPSEEK V4 灰度测试正式版,很出色,很厉害,很优秀,但不完美,还是会有BUG的,这是最真诚的话,且不接受反驳。
50 07-18 伊甸的亚托莉 【疑似deepseekv4正式版】ai里的贝多芬 多种音乐生成测试
51 07-18 -温醨- 疑似deepseekv4灰度测试,仅提供部分素材,另外部分素材由ai自己生成,逻辑一遍过,第二遍改了下画风
52 07-18 小小小名不是小明 V8声浪模拟,比地平线好听 DS V4 灰测生成
53 07-18 康康不说骚话 霓虹浪人-deepseek一锅出版本
54 07-18 鱼村长233 Deepseekv4正式版灰测,音乐创建和nes模拟器
55 07-18 雨下冰梧桐 单网页小游戏《方块世界》,这次我不蹭DEEPSEEK V4灰度测试正式版,抛弃流量密码,看一看真实数据。
56 07-18 kdzzzds dsv4灰度版 (ds自己建模) csgo 死斗模式+爆破
57 07-18 果汁盒Summer 用DeepSeek(疑似正式版)复刻战双帕弥什
58 07-18 stupid_scout DeepseekV4正式版 类战地游戏彻底翻新 加了许多内容(下载链接在评论区)
59 07-18 LikrFG 疑似 deepseek v4 正式版模型测试
60 07-18 小小小名不是小明 有V8!DS v4正式版游戏两则【马里奥 和 赛车】快速更新第四期,纯玩嗨了分享成果
61 07-18 添星_tianxing deepseek正式版,4块钱做的小游戏
62 07-18 离梦aajjkk DeepSeekv4pro正式版一句话生成4000小球测试,完美无bug(max模式),附带性能测试。
63 07-17 冬眠の松鼠_ DeepSeek V4正式版 三角洲行动和森林
64 07-17 小小小名不是小明 DS v4正式版一句话生成【截图翻译工具、精品独立游戏】快速更新第三期,纯玩嗨了分享成果
65 07-17 Qilin1132 当你让deepseek v4灰度正式版和预览版做同样的游戏时
66 07-17 梦回0 deepseek v4正式版,制作飞车类型游戏
67 07-17 风味光猫 [东方project]DeepSeek疑似灰测与预览版一句话生成弹幕游戏对比
68 07-17 kdzzzds dsv4灰度--(不搜素材纯自己建模)1轮出+1轮修bug的超高质量gta
69 07-17 Augenstern_-__- v4灰度正式版-两个前端版本的对比
70 07-17 没有设置一 deepseek把claude fable5开源了☝️
71 07-17 星光与雪 DeepSeek v4正式版多轮对话复刻《王国·两位君主》
72 07-17 黑韬 DeepseekV4正式版手搓的歌剧《卡门》动画
73 07-17 s就是这样s 疑似DS V4正式版灰度,被这效果吓哭,这是我看过《我的世界》+《无人深空》融合的最好的一个版本
74 07-17 kdzzzds 来看看v4灰测前端 隐藏的巨大潜力(AI生成的提示词)
75 07-17 DaVinci-2nd 没灰度到dsV4正式版或者灰度到坠机了的进来
76 07-17 kdzzzds 核弹之作-疑似dsv4正式版灰度-我的世界/星际拓荒 曲率方块
77 07-17 e4glet DeepSeek 3D Walkman播放器
78 07-17 猫伊cat 【Vibecoding】Claude和Deepseek合力做的可以弹奏的小太阳
79 07-17 一爽电台 马斯克锐评Kimi k3!我只想说 deepseek NO1
80 07-17 浪迹天涯人free deepseek正式版灰度做雷霆战机
81 07-17 kdzzzds 谁敢质疑d老师的审美-疑似dsv4灰度之只用SVG图作我的世界
82 07-17 -温醨- deepseek v4正式版灰度,我的世界+孢子,二次修改
83 07-17 星光与雪 DeepSeek v4正式版多轮对话生成光遇
84 07-17 kdzzzds 疑似dsv4灰度- 我的深空,在原提示词基础上加强美术提示词看看效果(没用联网)
85 07-17 爱尔奎德的死徒 DeepSeek v4灰度正式版+Z-image生成游戏,学院昆特牌
86 07-17 小小小名不是小明 测试第二期!DeepSeek V4正式版全方面深度测试 附带kimiK3
87 07-17 夜之叶归一 deepseek v4正式版 测试
88 07-17 夜之叶归一 deepseek v4正式版测试,一句生成游戏
89 07-17 -温醨- 疑似dsv4正式版灰测,我的世界+孢子
90 07-17 冬眠の松鼠_ DeepSeek V4 魔兽世界 高清重制版
91 07-17 Delight-linger 双日凌空! 对比测试K3与Deepseekv4(灰度正式版)实力
92 07-16 kdzzzds 不严谨测试
93 07-16 PotnQ DSV4正式版灰度无人深空超级伪无缝登陆
94 07-16 探归者夕迁 Deepseekv4正式版-我的太空计划 试玩 软着陆登月
95 07-16 冬眠の松鼠_ DeepSeek V4正式版 多轮对话生成魔兽世界
96 07-16 UPLUZ 【疑似DeepSeek正式版】像素鸟flappybird-预览版与正式版效果对比
97 07-16 stupid_scout DeepseekV4正式版 测试传送门(功能实现很不错!还有一些传送门问题测试)
98 07-16 黑韬 DeepseekV4正式版生成的逆转裁判!
99 07-16 洛必达实验室 Kimi K3 性能简直炸裂!又一个 DeepSeek 时刻要来了?人工智能AI
100 07-16 cacl2_lhg 疑似Deepseek V4正式版生成的仿马力欧赛车小游戏
101 07-16 Augenstern_-__- v4灰度正式版,svg拉弓动画测试
102 07-16 stupid_scout DeepseekV4正式版 我只是让他删注释 竟然顺手修了游戏bug
103 07-16 s就是这样s 疑似高维空间碎片,Deepseek正式版灰度,一句话生成《地平线6》
104 07-16 梦回0 deepseek v4 正式版生成只狼版网页游戏
105 07-16 stupid_scout DeepseekV4正式版生成的战地游戏(效果吊炸天了!!)
106 07-16 吧必须徐112 【7月最新】免费白嫖!DeepSeek V4 自由版免费用,不限次数!
107 07-16 三十丁尼 dsv4生成可自定义音乐的下落式音游
108 07-16 Delight-linger 核弹爆炸,dsv4正式版一次性生成多款安卓游戏,涵盖跑酷,AR,2D打击等等
109 07-16 UPLUZ 【疑似DeepSeek正式版】Besiege围攻,物理模拟很不错
110 07-16 飞越麦当劳 deepseek v4正式版灰测,生成恐怖游戏
111 07-16 飯綱九龍 【东方】Deepseek-v4正式版(疑似)灰度一次生成 - 东 方 智 械 录
112 07-16 Augenstern_-__- dsv4灰度正式版-修正前两期以及展示骑马与砍杀
113 07-16 Augenstern_-__- 疑似dsv4正式版灰度-后室
114 07-16 Augenstern_-__- 疑似dsv4正式版灰度-最有氛围的日系恐怖游戏,预览与正式版对比
115 07-15 UPLUZ 【疑似DeepSeek正式版】塞尔达世界
116 07-15 Delight-linger Deepseekv4正式版-坎巴拉太空计划+我的世界---实操登录月球(硬着陆)3轮对话
117 07-15 作手阿九 deepseekV4pro一句话生成的幸存者游戏
118 07-15 飯綱九龍 疑似dsv4正式版灰度-我的世界X无人深空
119 07-15 Delight-linger Deepseek正式版灰度 木筏求生+我的世界(我的木筏)
120 07-15 齐天大圣wy Deepseek 你这个死肥鱼到底干了什么
121 07-15 桶水test-2 被deepseek震惊了
122 07-15 kdzzzds 疑似deepseekv4正式版灰度 我的世界+无人深空(调小声音)
123 07-14 kdzzzds (以坠机)疑似dsv4正式版灰测 暮色森林(做一半坠机了)
124 07-14 kdzzzds 疑似deepseekv4正式版灰度测试-拳皇
125 07-14 kdzzzds 疑似deepseek v4正式版灰度-我的世界地狱末地地形测试(地形为一次出)
126 07-14 kdzzzds 疑似deepseekv4正式版灰测-无人深空
127 07-13 2258ppq dsv4 正式版 测试2
128 07-11 不可以包含特殊字符 DeepSeek4.0正式版灰度测试小游戏开发
129 07-10 飞越麦当劳 不知道是不是被deepseek v4正式版灰度到了,用一句话生成的,但效果确实好
130 07-09 冬眠の松鼠_ DeepSeek V4正式版一句话生成《后室》和恐怖游戏
131 07-09 小小小名不是小明 国产之光再进化!DeepSeek V4正式版简测,附带横评Gemini3.5 Pro / Grok4.5等
132 07-09 Delight-linger 灰度测试dsv4正式版?泰拉瑞亚-以撒结合-RTS-浏览器操作系统-割绳子-星露谷全测试
133 07-08 冬眠の松鼠_ 灰度测试:DeepSeek V4正式版一句话生成GTA5
134 07-07 冬眠の松鼠_ DeepSeek V4正式版一句话生成CSGO(生成记录透明)
135 07-07 冬眠の松鼠_ DeepSeek V4正式版生成的《我的世界》
136 07-06 System-Win-FUN Deepseek出现恶性漏洞,在回答问题的时候有几率会一直回答那个收到。

What 121 tests actually show

The data comes from the open dsv4ga-news-gather repository, which indexes public Bilibili posts and preserves titles, creator names, timestamps, descriptions, comments, shared conversations and download links. The Jul 19 snapshot breaks down like this:

121gray-test videos
53independent creators
40shared chat sessions
12day window (Jul 7–19)

Posts per day

Slow trickle until Jul 15, then a wall of releases peaking Jul 18.

Jul 6
1
Jul 7
2
Jul 8
1
Jul 9
3
Jul 10
1
Jul 11
1
Jul 13
1
Jul 14
4
Jul 15
8
Jul 16
23
Jul 17
29
Jul 18
40
Jul 19
22

Top 10 creators

Only kdzzzds and Delight-linger shared substantial session logs. Green = 5+ shared sessions, orange = 1–4, blue = none.

kdzzzds
14· 12 shared
小小小名不是小明
8· no shares
冬眠の松鼠_
8· no shares
UPLUZ
8· no shares
Delight-linger
7· 11 shared
Augenstern_-__-
6· 1 shared
夜之叶归一
4· no shares
stupid_scout
4· no shares
星光与雪
3· no shares
梦回0
3· no shares

What people actually tested

Categorised from 136 titles. Games dominate — production backends, security and long-horizon work barely appear.

Other game / prototype
4332%
Voxel / Minecraft
1712%
Card / RPG / arcade
1712%
Physics / space sim
129%
Shooter / military
118%
Music / audio
107%
Racing / open world
86%
News / roundup
75%
Horror
54%
Chat / vibe demo
32%
Tool / engine test
32%

Two biases matter more than any single number. Selection bias: most of these are "wish coding" tasks — browser games, 3D scenes, SVG animation, physics toys, music tools. The set is strong evidence for rapid prototyping and front-end generation; it's weak evidence for production backends, security, large-repo maintenance, or anything that takes months. Transparency bias: only 3 of the top 10 creators shared their prompts or chat logs. Treat video titles that say "official version" as marketing — the creator usually only suspects a gray route. We use suspected V4 gray build throughout.

What V4 actually seems good at

One-prompt apps now run, instead of just looking pretty

The strongest pattern is broad implementation from a thin spec. The set includes voxel worlds, shooters, rhythm games, survival games, three.js scenes, music generators and engineering visualisations. A representative GTA-style SVG demo was described as one main generation plus one bug-fix round, with a shared conversation attached. That's not "one sentence, one shipped game". It's "the model picks an architecture, wires enough of it together to demo the idea, and no longer hands back an empty mockup".

Strong 3D, SVG and lightweight simulation

Three.js scenes, custom SVG assets, game physics and interactive simulations repeat across the collection. An air-liquid CFD-style page reportedly used PBF for liquid and LBM for gas in two conversations. The creator also flagged unrealistic aerodynamics and the need for more repair — exactly the line between an impressive prototype and a trustworthy engineering tool.

Price may be the real headline

Community comparisons repeatedly describe V4 and Kimi K3 as close on practical projects. If the final API keeps DeepSeek's usual price advantage, "near-frontier capability at infrastructure pricing" matters more than winning any single head-to-head demo. See our AI API price comparison for current context.

Where V4 still struggles

  • Polish and taste. Kimi K3 is often preferred for front-end composition and long-form wording. V4's visuals can be impressive, but quality swings sharply with prompt design.
  • Instruction boundaries. One demo asked the model to delete comments; it also fixed a game bug on its own. That initiative feels magical in a toy, and is a review risk in a real repo. Review every diff — don't accept "it runs" as sufficient.
  • Long multi-turn reliability. Sessions still crash, lose direction, or accumulate architectural debt over many rounds. A strong first pass doesn't guarantee a strong twentieth pass.
  • Physics correctness. A convincing fluid or gravity simulation is not valid CFD or mechanics. Visually right ≠ numerically right.
  • Reproducibility. Gray routing looks inconsistent. Different users may have hit different models, tiers or sampling behaviour.
  • Evidence quality. Many videos omit the full prompt, model identifier, failure count, token use and manual edits.

The honest description is "high-ceiling prototype generator with real coding ability" — not "a senior engineer that no longer needs review".

V4 vs Kimi K3 vs GPT-5.6 vs Fable 5.0

There's no controlled four-model benchmark here. The table below is a workflow-oriented synthesis of the gray-test evidence and the user reports around it — not a leaderboard.

ModelLikely advantageWhere V4 standsBest role
DeepSeek V4Cost, broad implementation, rapid prototypesBaselineDefault daily coding and first implementation
Kimi K3Front-end polish, presentation, writingV4 is slightly less polished but often functionally comparableUI pass, product copy, visual refinement
GPT-5.6Review consistency, tool ecosystem, cross-file reasoningV4 may be the cheaper substitute for routine work; not enough evidence for a general winCode review, validation, stubborn bugs
Fable 5.0Long-horizon implementation, multi-turn recoveryV4 looks close in selected demos but less reliable at the edgeEscalation model for the hardest remaining work

V4 vs Kimi K3: the most credible direct material points to a close contest. K3 looks better on front-end and writing; V4 may be stronger value and competitive on implementation. One comparison video reached tens of thousands of views, but prompts and scoring weren't standardised.

V4 vs GPT-5.6: isolated comments claim wins and losses on particular web projects. These are test ideas, not benchmark results. The useful question is whether V4 can pass your repo's test suite at lower cost.

V4 vs Fable 5.0: impressive game demos make the comparison tempting, but the collection doesn't establish parity on large codebases, security review, or long autonomous runs. Treat Fable 5.0 as an escalation option until reproducible tests say otherwise.

A practical setup

The cheapest setup that doesn't bite you later:

  1. Let V4 do the first 80–90%. Planning, scaffolding, implementation, tests, docs, ordinary bug fixing.
  2. Verify locally. Lint, type-check, unit and integration tests, and a human diff review. Non-negotiable.
  3. Escalate with evidence. When stuck, send the diff, failing tests and a narrow question to Kimi K3, GPT-5.6 or Fable 5.0 — not the same vague prompt again.
  4. Keep one light premium plan, not several full subscriptions. Use it as a reviewer and recovery model, not the default token burner.
Recommended: V4 as the high-volume implementation model, plus one lightweight premium subscription for independent code review and problems that survive a few turns.

This only works if the second model gets evidence — diffs, logs, failing tests. Asking two models the same vague question usually doubles cost without improving reliability.

What to verify when V4 officially launches

Judge the official release on reproducibility, not launch-day demos. Check:

  • the exact API model ID, and whether web, app and API use the same tier;
  • context window, output limit, tool calling and structured-output support;
  • price per input, cached input and output token;
  • rate limits, peak-hour routing, and whether "Pro/Flash/Max" tiers behave differently;
  • repository-scale editing, test completion and instruction adherence;
  • how often identical prompts reproduce the gray-test quality;
  • licence, data retention and deployment options.

We'll update this page when DeepSeek publishes official model cards, API docs and pricing. Until then, claims about the final model stay provisional.

FAQ

Is the official version already available?

The collection documents suspected gray routing, not a confirmed release. A video titled "official version" is not official confirmation from DeepSeek.

How good is V4 for coding?

The evidence is strongest for rapid web prototypes, games, SVG, three.js and iterative bug fixing. It's thinnest for large production repos, security-sensitive code and long autonomous work.

Is V4 better than Kimi K3?

Not across every task. V4 appears close and may offer better value; Kimi K3 often has an edge in front-end polish and writing. Run the same prompt, runtime and acceptance tests before choosing.

Should I cancel my premium coding subscription?

For many users, downgrading to one lightweight premium plan makes more sense than cancelling everything. Let V4 handle volume and keep an independent model for review and escalation.

Where can I inspect the original gray tests?

Start with the GitHub index of all 121 videos. Representative cases: broad V4 + Kimi K3 test, 3D physics comparison, 8192-block Minecraft test, proactive bug-fix example.