DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis • Hacker News Board

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#41

Post by coder543 » Fri, Jul 31, 2026, 12:47 PM UTC

The weights were just released a few minutes ago: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#42

Post by WhitneyLand » Fri, Jul 31, 2026, 12:48 PM UTC

It’s exciting that a model scoring this high is dirt cheap.

It’s also so inefficient, when they release the full performance numbers it’s not going to be good.

One example, it takes about 3.6x more tokens to finish the same work as Gemini Flash 3.6.

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#43

Post by theanonymousone » Fri, Jul 31, 2026, 12:51 PM UTC

theanonymousone wrote: post #9

monooso wrote: post #3

404. I believe this is the correct URL: https://artificialanalysis.ai/models/deepseek-v4-flash

Yes, sorry, I went into anti-procrastination mode after I posted. I hope someone fixes it.

@dang is that how we call you?

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#44

Post by zawaideh » Fri, Jul 31, 2026, 12:52 PM UTC

SJMG wrote: post #30

avazhi wrote: post #22

Earlier quoted context omitted.

Western models censor just as much shit as the Chinese models do, big guy, it’s just different material. While we should be pushing for universal fully uncensored models, this comment is lazy and trite at this point. But you already know that.

Oh? What are the American model censorship tells?

Genocide in Gaza...

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#45

Post by theanonymousone » Fri, Jul 31, 2026, 12:53 PM UTC

WithinReason wrote: post #33

NooneAtAll3 wrote: post #24

what a horribly heavy and resource-consuming website...

we need a benchmark website benchmark

A benchmark website to benchmark benchmark websites?

Or a benchmark to benchmark benchmarks?

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#46

Post by storus » Fri, Jul 31, 2026, 12:54 PM UTC

I hope they somewhat fixed the hallucination and forgetting plagued V4 previews and that it wasn't just benchmaxxed but the numbers hold in reality. Then it would be my choice for 2x DGX Spark or 2x RTX Pro 6000.

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#47

Post by shalom1112 » Fri, Jul 31, 2026, 12:58 PM UTC

zawaideh wrote: post #44

SJMG wrote: post #30

Earlier quoted context omitted.

Oh? What are the American model censorship tells?

Genocide in Gaza...

[dead]

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#48

Post by WithinReason » Fri, Jul 31, 2026, 12:59 PM UTC

theanonymousone wrote: post #43

theanonymousone wrote: post #9

Earlier quoted context omitted.

Yes, sorry, I went into anti-procrastination mode after I posted. I hope someone fixes it.

@dang is that how we call you?

no, like this: hn@ycombinator.com

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#49

Post by hxii » Fri, Jul 31, 2026, 1:00 PM UTC

I’m wondering if they did anything to address the DSML tool calls leaking. Has been an issue with both Flash and Pro so far.

Re: DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

#50

Post by Der_Einzige » Fri, Jul 31, 2026, 1:01 PM UTC

Daily reminder that none of these numbers are valid in a world where no one publishes the sampling settings used.

Daily reminder that improving your samplers from the garbage default top_p/top_k to min_p or subsequent methods dramatically improves the performance of these models, and makes most quantities like measured "verbosity" and subsequent calculations of "intelligence per token" meaningless

Daily reminder that no one, including within academic AI research, AI engineers, normies, etc takes LLM sampling seriously enough.