DeepSeek
DeepSeek

DeepSeek V4 Pro

TL;DR

Users celebrate the Technical Coding performance and aggressive Pricing of DeepSeek V4 Pro, but significant frustration persists regarding extreme Inference Latency and potential Geopolitical Bias within the model's outputs.

YouTubeRedditHacker News
555 comments analyzedApr 29, 2026

Popularity metrics

Last 30 days · 25 items analyzed
  • Related views
    9.5M
  • Related endorsements
    164.7k
  • Related comments
    10.1k
  • Buzz
    Rising fast

Comment distribution

  • Overall
    52%
    48%
  • YouTube
    50%
    50%
  • Reddit
    54%
    46%
  • Hacker News
    53%
    47%
PositiveNegative

Comment summary

Technical Coding and Programming Performance

290 comments

Developers report superb results in bug detection and scientific coding, with many considering it a top-tier tool for technical tasks.

Pricing and Open Weights Accessibility

79 comments

The model is widely praised for its disruptive cost efficiency and the availability of open weights compared to closed-source rivals.

Inference Latency and Hardware Barriers

94 comments

Significant complaints focus on 'next level' slowness during reasoning tasks and the massive VRAM requirements for local hardware execution.

Geopolitical Concerns and Content Bias

92 comments

Skeptics highlight potential censorship and political biases tied to the model's Chinese origin, alongside general data privacy anxieties.

Top comments

  • Claude seems the most realistic, but Gemini seems like it best understood the assignment.
    YouTube@ttrestle19,956View original
  • You asked for fractal though, which clearly is more like gemini and deepseek's result
    YouTube@squaidinkarts212View original
  • Competition is good. Open Source even better. Cheaper and good-enough AI should benefit everybody. Thank you Openseek 🙏
    YouTube@FlemmingNLarsen174View original
  • To me the v4 pro seems to be hugely undertrained. I expect we're going to see huge gains in that model when we get new checkpoints in coming months.
    RedditPuzzleheaded-Drama-8152View original
  • Sempre que um bom modelo oriental é lançado eu comemoro. Ter alternativas ao oligopólio do império do norte é fundamental
    YouTube@DerSteppenwolf1979125View original
  • Qwen 3.6 27b. Melhor modelo que eu rodei localmente ( o que é esperado de um modelo novo da qwen), mas por uma margem insana. Eu não uso mais api pra nada que não seja os projetos mais ambiciosos. Alteração no código, navegação no pc, pesquisas, e até exploração mesmo de temas. Eu nunca fiquei entusiasmado pra usar um modelo por tanto tempo depois de ter testado. Finalmente é um modelo que é BOM O SUFICIENTE
    YouTube@ulisses17786View original
  • O deepseek tá salvando minha vida. Tava cogitando um plano max do claude e fazendo as contas pra resolver um projeto gráfico em c++. Ele simplesmente mudo completamente o jogo. Vc trabalha lindamente com ele. Pau no c* dos EUA e sua arrogância. Sempre querendo f*der com o consumidor. Se tem algo que entrega valor pra humanidade em IA hoje, que torna acessível, são os modelos chineses. DeepSeek tá dando aula!!! inclusive pra outros modelos chineses que se aproveitam da reserva de mercado que os EUA impõem!! e parabéns pelo vídeo e pelas análises!! não aguentava mais essas análises de oneshot inúteis!!!
    YouTube@vitorgrilovital77View original
  • i use v4 pro and flash via openrouter as worker subagents. so they dont design, plan, discuss or research don't do anything other than implementations. u get gpt 5.4 level performance almost for 5.4 nano pricing, in fact its even cheaper than nano. flash is even crazier. i dont think there is any model out there which can compete with this in cost effectiveness. its by far the cheapest sota model. like by far. the reason I dont fully use them is becuz i still prefer claude/gpt models for design etc, not that i tested chinese ones yet on that front. i probably should.
    Reddit2Norn64View original
  • The funny thing is that I use both ChatGPT and DeepSeek in my personal life, but they're for different purposes, based on the results I've interpreted. This video fully answered why I prefer them for their respective purposes: 1. DeepSeek for anything technical or objective (e.g. troubleshooting my PC OS, mathematics, science, etc.) 2. ChatGPT for anything social or subjective (e.g. social planning, restaurant recommendations, arts, etc.)
    YouTube@eddiestilll60View original
  • Yeah though that's for dense. MoE have different scaling, depending on sparsity - https://arxiv.org/abs/2507.17702
    RedditFullOf_Bad_Ideas72View original

Source breakdown

Graph based on sampled comments per item (n≤30)