deepseek v4.1 flash vs secret model tested both with the same prompt at the highest reasoning available > v4.1 flash took 2 hours to finish both tests and cost $2.60 > secret model finished both in 30 minutes, can’t reveal cost yet v4.1 flash spent a lot of time... overthinking and running random tests, which slowed it down thanks AIHubmix for providing api access so i could test v4.1 flash hoping that team will cut down the overthinking before the full release, but for now here are the results which model did better here?show more

voxel pagoda test for both models

Bro how u use deepseek v4.1 flash in Ai hub mix last time I used 😒 hit the limit without even building something Are u using paid or free lmk?

i have credits, load like 5 to 10$ and you should be good for testing almost all chinese models

2 hours is wild. but still deepseek v4.1 flash looks solid

Agreed but missed alot stuff

And The secret model is (IMO) next Gemini flash model. Only Gemini have true flash Decode speed. Before i thought that's Grok 4.7, but quality is not good for Frontier model, and elon said that is slower than 4.6, but better overall

hehe

Deepseek 4.1 flash Has better Water detail, and i think overall deepseek is better

@notjazii Always the trade-off between speed and depth. Hit this wall when testing models for my own stuff. A balance is crucial!

having the same feeling as when Astra launched, might as well test this out myself

deepseek goated bro, love it

Cooking that well in 30 mins is really impressive legend 👀

high feels less overthinking

wait fr? I tested on max in Hermes so lemme try with minimax test

yeah try opencode

not working in opencode no idea why

need to add custom model

so when can we know the secret model's cost

maybe in few days

Secret model is flash or pro?

it's fast fast I can't say flash or pro or air or lite or GPT 6-7

Secret model is way better hoping that it costed less

What is the secret model?

it's obvious grok 4.7 better

😂😭

ig we're cooked with Grok 4.7

what's the secret model tho? Qwen?

highest reasoning setting does that, it keeps re-checking an answer it already had.

ngl DS one looks more detailed

left one is definitely better

yeah it did better but some details are missing 100%

Is deepseek good at coding? Also what's the cheapest way to get access? 😭😭

nothing for now soon it'll be opencode but wait for now

this is why speed matters so much for agent workflows. 4x faster changes the whole experience

This secret model seems really decent at 3D. Could you prompt some other stuff? Could be simple. There’s a thing I like to test in every model - I ask for a three js representation of a temperature and umidity reader using a esp32. So far Astra is winning, but still quite bugged

I ran 3d ship tests so I'll post it later tonight

try high thinking , 4.1v better high

secret hmm

Did the extra 90 minutes actually translate to a cleaner pagoda, or was it just burning tokens stuck in thinking loops?

Impressive speed upgrade, that’s a game changer in ai models of today

dsv4 looks better imo but 2 hrs is a lot btw jazi, how about not labelling the output next time and revealing in the comment instead, less bias and more fun 🙃

deepseek clearly better output. but time.....

Overthinking can tank efficiency. Speed without depth often leads to shallow results.

which do you prefer?

I like both, as one did faster and got good details and other did way better

降价了

Its behavior appear to be different for language used for prompting. I don't see over think issue when poking (戳戳) in Chinese.