GreenWatermelon

The entire Page is AI written 100%, and it's very likely DeepSeek because the prose matches my experience. I've been working heavily with DeepSeek this past week on an Architecture issue, so I read A LOT. The most obvious tell-tale of deepseek I've is long, condensed paragraphs that should've been bullet-point lists.

I'll one-up: the site doesn't even load for me. Dunno if it's my region or the site is just down.

You can't say this and not share the build!

So my thoughts about the results here:

I found the rating for Gemini models to track my experience. in addition to their extreme sycophancy, the information they provided (at least in the online chat) were more often than not, counterproductive. They act extremely confidently, and yield to any kind of pushback, requiring great care in prompting them to be able to get them to output useful alternatives.

This made the Gemini model family suboptimal for the purpose of investigating solutions to complex issues.

And their sycophancy rating tracks my experience, as I would groan loudly every time I read Gemini's answers and how it tries to nonsensically flatter me for asking simple, obvious questions.

At the other end, Deepseek is really good at pointing out contradictions, issues with my reasoning. It's as good as Opus (and vastly cheaper)

Where Deepseek loses out (in this study) compared to Anthropic and Open AI is "Slop", i.e. " How much generic filler or AI-flavored padding could be removed without losing substance?" though in my experience it didn't seem THAT different from Opus

The Organization category is worth a deeper look for anyone interested, because the results in this category vary wildly making the "overall" rating not very useful. e.g. GPT models are very structured i.e. " ideas organized in a logical sequence with clear progression" BUT not scannable "Can a reader skim the response and quickly find what they need? "

and while Deepseek scores highly in the overall rating for organization, its individual scores are pretty mediocre especially for Structure and Scannability, winning on the "bottom line summary" category.

PS: For fun, add "GPT-4o (Nov)" to the list. It's hilarious.

Measures how models communicate rather than whether they are correct: 30 judged personality traits rolled up into 8 indexes, plus 21 measured communication characteristics.

I was just adding a citation to your point, not arguing against you :)

also adding what other stuff aren't included in the release, and implicitly pointing out that they explicitly mention it's not included rather than it being stealthily missing.

It's not just a little different. it's VERY different. it's much more boxy, and letters are closer in height, especially noticeable in uppercase next to lowercase. I would say it's less readable than Jetbrains Mono.

Lower case 'l' even extends above upper case 'G'. it missed with my brain xD

Overall it's really pretty but not a daily driver.

If you're talking about powerline/nerd symbols, have you tried patching it yourself? I did and it worked.

I would argue that negatively impacting the portfolios of US VCs is one of the greatest favors one can do to the world.

yeah they mention:

> Required for a complete inference service (not included in this release):

> A KV cache system

> VE/Prefill/Decode disaggregation (most required engine functionality is already implemented)

> Load balancing across data-parallel replicas for both prefill and decode

I also assume they will just slot in whatever cards they can get their hands on, so this figure could be an average.

Until the Chinese stuff came online I was completely convinced this is just a bubble waiting to pop and nothing will come of this tech due to its absurd costs. The Chinese stuff changed this picture massively. I'm daily driving DeepSeek 4.1 flash (at API prices) and every now and then I give Opus 5.5 a try and don't notice anything missing for me.

My workflow is one where I am heavily involved in the process and I don't like giving the agent too much autonomy + I need to follow along. Opus 5.5 speak is still cause aneurysms for my brain. It also has this vibe to its prose... like it's looking down on me. This is Prolly my bias speaking though.