Cristian

If @uxwizz was so loved by customers 5-10 years ago, imagine how good it is today after years of extra work :)

Remember to change the thermal paste on your CPU once every few years.

@johnstejskal Hey, I remember seeing some video of this game a while back. When will it be live?

I have a confession. I'm always scared to share footage from my game. 5+ years solo dev will do that i guess... So to fight this fear, here’s 5 minutes of raw gameplay from one of the first levels. No edits. No hiding!! The game's called Blood And Mead, you can wishlist it on steam ( Link in reply⬇️) #gamedev #indiegame

@Mathgeek007 @thehorizon2b2t Big number first always makes it more exciting: $10k watch vs $1 watch vs $1 watch vs $10k watch You are already curious to see what a "$10k watch" looks like.

@thehorizon2b2t It establishes protagonist vs antag - when you write "X vs Y", you're sort of implying X is *taking on* Y, as in X is the aggressors. Nobody wants to see a story of 10 pros slaughtering 1000 newbies, but newbies rising up against the strong people is compelling!

Insane how the CTR goes up by 15% if you just switch the numbers around in the title

So MiMo-V2-Pro is slightly worse than GLM 5, but 2x faster on average.

There's something odd about the Hunter Alpha -> MiMo V2 Pro release. The no-reasoning version gets the same score for both the stealth version and live version. But the medium reasoning version performs much better for MiMo V2 Pro than the stealth Hunter Alpha. Looking at the results, it looks like they changed how V2 Pro medium reasoning works, so it reasons for longer (27k tokens vs 18k tokens), leading to better results, comparable to GPT-5.4 almost.

@jonatanpallesen Or everyone is getting smarter... IQ 100 = average, and then everything else is relative to that.

The proportion of high-IQ people in the world is declining fast.

@AdamDraper > @ProtonMail, rejects about 1.2 million mis-addressed e-mails per week to the gail.com domain. Lol

accidently typed in gail.com instead of gmail, and its my new favorite website.

@thegenioo Neah, mimo-v2-pro (hunter alpha) is a lot better: aibenchy.com/compare/minima…

Idk how but MiniMax M2.7 is a better model on most benchmarks than Xiaomi MiMo-V2-Pro while being marginally cheaper too I do see that MiMo-V2-Pro gives better UI sometimes but not as reliable overall as M2.7 in coding

@JasonBotterill In my tests, grok 4.20 does around 10-20% better, but it's 2x more expensive: aibenchy.com/compare/openai…

Lmao even the mini version of GPT-5.4 outperforms Grok 4.20

@slow_developer gpt 5.4 is yapping so much, it won't be too hard to optimize it

how is this even possible? gpt-5.4 pro is using far fewer tokens and costing much less overall than gpt-5.4 xhigh either this is a mistake, or openai discovered an efficiency paradigm -- which could simply be a good system that cleans up training data to only the high-quality stuff

@kodee2k @marmaduke091 I don't think it was Hunter Alpha, the benchmarks for it are quite different than m2.7

@XCSme @marmaduke091 It was and its really bad. It is book smart for very specific tasks but for actual reasoning its pretty bad in my benchmarks it gets smoked by even GPT 5.4 nano

MiniMax M2.7 is out on web! Check it out here: agent.minimax.io

MiniMax 2.7 is performing worse than MiniMax 2.5... They fixed the problem with huge reasoning tokens usage though.

@adonis_singh Lol, as a dev that struggles with deciding what version to bump, this is very accurate. Sometimes even the middle, Default version is a shame version, but the change/fix is too major to be buried in 2 decimals

2 decimal points is actually insane, reminds me of this

I am using @AIBenchy almost every day. I also added the OpenRouter input/output prices, just to save myself some time from having to check it there. Today I was looking into which model should I use for translations, I decided to switch from qwen 3.5 plus to gemini 3.1 flash lite, as they cost the same, perform the same, but gemini 3.1 flash lite is like 10x faster.

@corbin_braun 📊📈🔥👀🆎️🐞🔐

pitch me your startup with 0 words.

@dejavucoder Sonnet and Opus are a lot closer than gpt 5.4 and gpt 5.4 mini

is what opus 4.6 to sonnet 4.6 the same as gpt 5.4 to gpt 5.4-mini? we will find out. so far this hasnt beem true

@JasonBotterill Sort of, it is 5x cheaper, but also considerably more inconsistent in the correctness of the output. aibenchy.com/compare/anthro…

5.4-mini is roughly Sonnet 4.6 intelligence but 70% cheaper and like 3x faster

@Mayhem4Markets They lose money on each API call, but make it up in volume.

OpenAI is an actual cash furnace. Just setting money on fire. 🔥💵🔥 The company is expected to burn through $68M per day in 2026. That figure rises to a staggering -$156M per day in 2027. Then -$200M per day by 2028. This sets up for a cash burn of $665B by 2030. No wonder they're pivoting to rent compute rather than building data centers. Who would want to invest in this? What is the path to profitability? Where do investors ultimately draw the line if RoI is elusive?

@cryptopunk7213 5.4 mini is a lot weaker than 5.4, it makes a lot of mistakes/is not consistent enough imo:

f*ck yes openai just dropped 2 new models thats 2X faster and cheaper but AS GOOD as gpt-5.4 (it actually beats it imo) - $0.20 PER MILLION tokens is practically free lol (nano) - matches 5.4 at coding, computer use, reasoning but much much faster. - oAI designed these models to WORK WITH 5.4 orchestrating them. tell 5.4 to delegate work to these models. sub-agent architecture tells me they're really leaning into multi-agent models. openai team are now shipping new models every month or so, pretty dope