DeepSeek V4 – almost on the frontier • Hacker News Board

Re: DeepSeek V4 – almost on the frontier

#141

Post by soerxpso » Sat, May 02, 2026, 2:26 PM UTC

eckelhesten wrote: post #67

dbeley wrote: post #45

Earlier quoted context omitted.

The cool thing about open-weights model is that you are free to use alternative providers that won't phone home to the original model creators. I see 6 alternative providers listed on Openrouter for DeepSeek V4 Pro for example.

At least that’s what they’re telling you. It’s a ”trust me bro” scenario. I’d rather use the phone home version (deepseeks own endpoint). The benefit is that I’m fairly certain that they actually host the model I’m paying for.

Some providers are based in the US or EU and would face legal repercussions for lying about what they do with your data. It's a bit more than "trust me bro". Off the top of my head, you can use Fireworks, for example, which is based in California and would face the same consequences for lying about their data policy as OpenAI or Anthropic would.

Re: DeepSeek V4 – almost on the frontier

#142

Post by alex1138 » Sat, May 02, 2026, 2:27 PM UTC

Does it censor mentions of what happened in Tiananmen Square in 1989?

Re: DeepSeek V4 – almost on the frontier

#143

Post by baldai » Sat, May 02, 2026, 2:29 PM UTC

gmerc wrote: post #44

baldai wrote: post #31

Earlier quoted context omitted.

Only similarity it has to Opus 4.6 is the 4 in the name. I do not understand these dishonest comparisons. OOS models are vool, cheap and promising for a future -- but why are we pretending they are better than they are?

Speak for yourself. I found switching from Opus 4.7 to be completely painless and in fact, due to the reliability of Anthropic’s API, less of a friction despite slower response times. Zero issues on a large mono repro

Hi, I am happy it works well for you. For me personally I struggle finding good use-cases in general for these OOS models. I am lightly technical but I do not manually code. So my flow is /grill-me (can take hours), make plan, review plan with 2. model, implement, review after implementation.

Maybe it is because my tasks are usually chunkier, or because I cant code myself that I struggle using cheaper models. Feels like at every stage of this process SOTA model improves it by 5%, which adds up.

But I am maybe ignorant of Opus level. My main driver is 5.5 and Opus is there for frontend and 2. opinion. In a past I also used Claude models for the chatting phase, but 5.5 took over recently. Maybe Deepseek is closer to Opus and I just overestimated the model compared to 5.5? I tried to give it benefit of being similar.

Recently I started experimenting with Deepseek Flash, maybe hoping if plan is solid enough it can implement quickly and cheaply, but for now it feels not worth it.

How do you use the model to see the benefits? Have you tried 5.5 and can you compare to that one as well?

Thanks.

Re: DeepSeek V4 – almost on the frontier

#144

Post by singpolyma3 » Sat, May 02, 2026, 2:32 PM UTC

ozgrakkurt wrote: post #24

prism56 wrote: post #18

Earlier quoted context omitted.

If the data is opensource on github, then in my opinion it should be fair game.

IMO this is unfair for GPL or similarly licensed code. Seems ok for MIT like licensed code though

There's no difference. Either you need to follow the license or you don't. MIT has requirements still.

Re: DeepSeek V4 – almost on the frontier

#145

Post by singpolyma3 » Sat, May 02, 2026, 2:33 PM UTC

edg5000 wrote: post #54

ozgrakkurt wrote: post #24

Earlier quoted context omitted.

IMO this is unfair for GPL or similarly licensed code. Seems ok for MIT like licensed code though

I think AI will create an open source dark age. Gradually, we'll see a lot less new good open source code. A gradual shift back to the proprietary world. Simmilar to the 1950-1990 period.

Why would giving more people software freedom and the ability to reverse engineer nonfree code result in a dark age?

Re: DeepSeek V4 – almost on the frontier

#146

Post by segmondy » Sat, May 02, 2026, 2:34 PM UTC

raincole wrote: post #35

The V3/R1 time and now are in such contrast. V3/R1 were hyped hard and barely usable for coding. V4 is much less hyped but (anecdotally) it has completely demolished all the Flash/Lite/Spark models.

They were and are still great for coding. They were not trained for agentic workflow and coding harness.

Re: DeepSeek V4 – almost on the frontier

#147

Post by pedrosorio » Sat, May 02, 2026, 2:34 PM UTC

cassianoleal wrote: post #75

cheshire_cat wrote: post #69

While the cost are lower than frontier models there are two factors that make DS4 Pro and K2.6 not as cheap as they might look. For DS4 Pro there's a discount going on for the official API, which sometimes gets overlooked and mixed up in discussions. Simon uses the full price in the comparison, so that's not an issue here. The other issue is that DS4 Pro and K2.6 often use way more reasoning tokens than the frontier…

Sure that can happen but it hasn’t been my experience. I just spent a whole day using it for some pretty hefty refactors, many rounds of back-and-forths, thousands of lines of code changes, reviews, investigations, many subagents running parallel tasks, the works. Total cost $0.95, altogether. I had attempted this with Opus 4.6 in the past and it burned through the $10 budget I’d given it before it returned from my i…

What harness do you use?

Re: DeepSeek V4 – almost on the frontier

#148

Post by singpolyma3 » Sat, May 02, 2026, 2:34 PM UTC

notrealyme123 wrote: post #34

prism56 wrote: post #18

Earlier quoted context omitted.

If the data is opensource on github, then in my opinion it should be fair game.

Things being public should not be enough. just because someone leaked your medical information to the public via a data breach should not make it fair game. There should be some rules.

There are rules. I believe that search engine indexing follows these rules and that so called "training" is search engine indexing.

But a court may differ in the future.

Re: DeepSeek V4 – almost on the frontier

#149

Post by segmondy » Sat, May 02, 2026, 2:35 PM UTC

edg5000 wrote: post #56

Has anybody used V4 hard, for the most challenging tasks (agentically, locally)? It's so hard to compare without putting serious time in it. Like spending a year daily with the model.

That's what an evaluation dataset is for, create your own and you can bench a model in a few hours to see if it fits your needs.

Re: DeepSeek V4 – almost on the frontier

#150

Post by ignoramous » Sat, May 02, 2026, 2:35 PM UTC

cedws wrote: post #113

The biggest differentiator for me: DeepSeek just does what I ask. I've tried using both GPT and Claude for reverse engineering recently, both refused. I even got a warning on my OpenAI account.

> even got a warning on my OpenAI account

Edit: https://chatgpt.com/cyber