Qwen 3.7 Max a prova: gran codi web, errors en 3D
Qwen 3.7 Max Preview sorprèn amb un joc de metro i una web de pasta, però falla en orientació i renderització. Què revela la prova i què ha canviat?
Una previsualització amb poca documentació inicial
Bijan Bowen prova Qwen 3.7 Max Preview el 19 de maig del 2026, quan el model acaba d'aparèixer al web de Qwen però encara no respon amb un identificador públic estable a l'API. Per explicar-lo, el creador recorre a la documentació de la generació anterior i evita atribuir al nou model una finestra de context no confirmada.
Aquesta prudència és important. Max és la variant allotjada més capaç de la família Qwen i històricament no s'ha publicat amb pesos oberts. Que altres models Qwen siguin oberts no permet descriure Max com a programari descarregable.
La situació va canviar immediatament després del vídeo. Alibaba va presentar oficialment Qwen 3.7 Max el 20 de maig i Model Studio va incorporar qwen3.7-max i la instantània qwen3.7-max-2026-05-20. La documentació actual el tracta com un model propietari per a tasques amb agents, codi i execució llarga.
El sistema operatiu web queda curt però té veu
El primer prompt demana un sistema operatiu fictici al navegador. El model genera només 672 línies, una resposta molt més curta que en altres proves del canal. Diverses aplicacions són decoratives i no hi ha gaire profunditat, de manera que Bowen el considera el resultat menys impressionant del conjunt.
Hi ha, però, dues idees bones. El control per veu obre aplicacions i respon en àudio, cosa que aporta una capa d'accessibilitat. El fons també pot produir gradients aleatoris fora de la selecció predefinida. Són funcions concretes que donen vida a un prototip modest.
La lliçó és que menys codi no sempre significa més eficiència. Si el prompt exigeix moltes interaccions, una resposta curta pot haver omès requisits. La validació necessita una llista de comportaments, no només obrir la pàgina i comprovar que sembla un escriptori.
Un joc de metro que eleva la prova
El repte del metro combina una escena estàtica amb un joc de trets en primera persona. Qwen produeix bancs, rètols, una llanterna, enemics amb cares, casquets i marques dels impactes sobre l'entorn. També incorpora so i un bucle jugable funcional.
Bowen el situa entre els millors resultats que ha vist amb aquest prompt, comparable en alguns detalls a models molt més cars. La llanterna i els impactes són especialment rellevants perquè depenen d'estat: no n'hi ha prou amb dibuixar una captura, cal que la interacció modifiqui l'escena.
No sabem si una segona execució hauria mantingut la qualitat. Un únic resultat excel·lent pot ser una mostra afortunada. Per comparar models caldrien diverses repeticions, mateixa temperatura, mateix entorn i una rúbrica que puntués cada requisit.
C++: primer falla, després millora molt
El joc de monopatí en C++ revela el valor de la iteració. La primera versió compila, però moltes superfícies no es renderitzen bé i els carrils no funcionen com cal. El model havia triat una solució orientada a Windows malgrat que l'entorn de prova era Linux, una decisió que complica la portabilitat.
Al final, Bowen torna al projecte amb una descripció breu dels defectes. La nova versió corregeix el renderitzat i transforma una demo decebedora en un joc visualment coherent, encara que els vianants i algunes animacions continuen estranys.
Aquest canvi és més informatiu que una generació perfecta. Mostra que Qwen pot localitzar i reparar una part del problema quan rep evidència concreta. També recorda que el cost real d'un agent inclou els intents de compilació, la inspecció humana i les correccions.
Vol, imprimeix i de vegades gira malament
El simulador de combat aeri presenta tres avions, terreny amb col·lisions i una física de vol prou utilitzable. Les bales deixen traça i una maniobra en espiral resulta convincent. Els enemics, en canvi, no semblen capaços de fer gaire dany i falta profunditat de joc.
La impressora 3D construeix formes per capes i s'assembla a un vídeo accelerat real. Algunes peces estan desplaçades o orientades de manera estranya. La bateria virtual pateix un problema semblant: sona i executa quatre patrons, però el bombo apareix girat.
Bowen ja havia observat errors espacials en altres Qwen Max. La sessió no prova una causa interna, però sí un patró visible: el model pot generar tots els components i, alhora, equivocar-se en la relació geomètrica entre ells.
Dues webs, dos resultats oposats
La pàgina de Sandglass, un laboratori creatiu fictici, compleix l'efecte de sorra però cau en una estètica tipogràfica genèrica que el presentador rebutja. És funcional i interactiva, però no transmet la identitat premium demanada.
La web de la salsa Ravioli Rosso és l'oposat. Qwen converteix el fons en un univers de tomàquet, oli, alfàbrega i raviolis, afegeix partícules al clic i construeix targetes amb moviment. El resultat és absurd i comercial alhora, just el to que buscava el prompt.
Els dos exemples mostren que «sap fer front-end» és una conclusió massa ampla. Pot executar mecàniques de canvas i composició, però la direcció artística depèn molt del prompt, dels patrons apresos i de la selecció posterior d'un humà.
Què confirma Alibaba després del vídeo
Alibaba descriu Qwen 3.7 Max com una base per a agents de llarga durada. En una demostració interna, afirma que va treballar 35 hores, va fer més de mil crides d'eina i va optimitzar un nucli per al xip Zhenwu M890. És una afirmació del fabricant i no substitueix una reproducció externa.
Model Studio ofereix modes de pensament i no pensament, cerca web i un context d'entrada de fins a un milió de tokens en serveis compatibles. Els preus i els descomptes han canviat des del llançament, per això convé consultar la taula oficial abans d'estimar costos.
La prova de Bowen es va fer al xat web, no en un arnès amb accés complet al sistema. Això limita la comparació amb agents que poden executar tests de forma autònoma durant hores, però també mostra què obtenia un usuari normal el primer dia.
Molt potencial, cap veredicte universal
Qwen 3.7 Max destaca en el joc de metro, la web de raviolis i la reparació del joc C++. És menys fiable en orientació tridimensional, en el primer sistema operatiu i en alguns simuladors. Aquesta barreja és coherent amb un model potent que encara necessita proves.
El vídeo no determina si supera tots els competidors, ni si Qwen 3.7 Plus tindrà una variant oberta equivalent. Sí que documenta un salt útil: el model passa de generar una maqueta a sostenir interaccions, so, col·lisions i una ronda de depuració.
Per a producció, la pregunta no és quin resultat fa més bona impressió, sinó quin percentatge de requisits passa de manera repetible, quant costa corregir la resta i si l'API garanteix la versió. Una demo brillant pot obrir el projecte; les proves decideixen si el poden tancar.
Contrast i context
Fonts consultades
-
01
YouTube — Bijan Bowen Qwen3.7 Max First Test – Hands-On With Alibaba’s SMARTEST Model!
-
02
YouTube Canal de Bijan Bowen
- 03
-
04
Alibaba Cloud Models Qwen compatibles amb cerca web
-
05
Alibaba Cloud Preus i instantània actual de Qwen 3.7 Max
-
06
Alibaba Cloud Model Studio Fitxa de Qwen 3.7 Max al catàleg de models
-
07
Alibaba Cloud Modes i cost del model dins el servei PAI
- 08
Font de treball
Transcripció amb marques de temps
Consulta la transcripció
-
0:00
, obre el vídeo en una pestanya nova
I mean, let's just look at this as a scene. Oh, it does have that. I've not seen that since the Opus 4.6. This is good. This is fantastic. Today, we're going to be taking a look at the newly released Qwen 3.7 Max preview model. And this is somewhat of a perplexing introduction, at least for myself, because this model seemingly appeared out of nowhere. There is currently no official announcement blog post or anything like that that I can actually reference in this introduction. The only mention of this is quite sparse and listed on the X profile for Alibaba Qwen, where you can also follow me. You can see here in the bottom left. Also, do feel free to subscribe, as I do want my 100K plaque. Um the followers here, I'm not as concerned about, but the YouTube subs, that's where it's at. So, with that though, I do want to make note that the only mention right here of Qwen 3.7 Max preview is right here in this post, where they mention it scores 13th place on the Text Arena leaderboard, which, you know, it's like they give trophies for everything nowadays. No, it's like the meme for that. But, excitingly though, they do also mention the Qwen 3.7 plus preview model. So, the plus model, at least previously for Qwen 3.5, also had an analogous open-source version to it that was the same parameter count. It is hopeful, fingers crossed, that the plus model for Qwen 3.7 will also give us some large 3.7 open-source model as well. There have been some rumors or speakings about open-source 3.7 models, which would be quite exciting. So, really, because there's not a lot to talk about here, I did try a sneaky little trick, which would just be to change the number in the blog post for the 3.6 Max preview page. Unfortunately, when we did that for 3.7, nothing shows up. I did also check specifically in the model studio and the API docs and everything like that. I did try just doing some API calls with the model identifier changed to Qwen 3.7 Max preview. And unfortunately, it's still not like fully available. But, on the website it is, so that's where we'll be testing it for today's video. Now, just for some pertinent information about the Qwen Max series, this is, as they say right here, the largest and most capable variant in the Qwen model family. Historically, the Max model has never been open weight, so this is something that is kept closed source and is, listed as they say here, the most powerful and most potent Qwen model. Now, I cannot definitively estimate that these will be the same exact specifications for the 37 preview model without actually being able to see it,
-
2:27
, obre el vídeo en una pestanya nova
but we can see that 3.6 Max preview did have a maximum output length of It was 64K, as we see right here, and a context window of 256K. So, with that, that's really small bit of information I can give for the specific model. And, as we can see right there, I had also tried that little trick where I changed it to 3.7 Max preview, and if I press enter right here, it would just give me some error pop-up and keep me on this specific page or something of the sort, but I digress. So, we're now going to get started with our tried and true browser OS test V2.5. Now, I was a little worried that just because it showed up here doesn't mean it was going to guaranteed work. Now, I'm very happy to see that the thought process has begun. That was a very, very fast chain of thought comparatively to what I've seen previously. I don't fully have recollection, but I did test the Max preview previous version like a month ago, and I seem to recall it thought longer than that, but I can't say definitively. Regardless though, what I was going to say was I was a little worried that it was just showing up here, but wasn't actually going to be live since it's not with the model identifier through API. Fortunately, it is live here. Seems to be um producing code at a rapid pace, so we'll see what we get. All right, the result is complete here, and I do have a couple of things to say before we even look at it. One, top-down 3D-ish. Okay, not necessarily what I'd said. Additionally to that, this is an extremely short script at 672 lines of code. I don't think I've ever had a result this short for this specific prompt. Okay, well, it starts out good. I guess I can say that. We can close that now. All right. First and foremost, nice. It does have a right click, which makes me quite warm and fuzzy inside. Now, we do also have the time in the bottom right as well as the date in my locale, as well as inevitably what is listed as that special feature, which is some voice command thing, and it even has an accessibility tool tip that pops up there. So, I do like that. I mean, the meantime, let's start with start. Okay, simple. I don't see any search feature here, which sometimes the models do go above and beyond and include. Although, again, the argument could be made that I'm not telling them to, so acceptable. Let's try now our GTA clone. You know what? That's Don't get me wrong. I'm infuriated because it's not 3D, and it even went ahead and called
-
4:41
, obre el vídeo en una pestanya nova
this 3D-ish. No, there's no ish to this. It's It's genuinely 2D. But, okay, so we can like shoot, and the arrow or the location of the Oh, okay. Location of the cursor shows our like that aims the weapon. Unfortunately, the car logic doesn't seem to work. As I press E and it shifts our position, but we don't actually get to take control of the vehicle. Let me just find one more and try it. Is this procedural? Uh okay, no, it is not. That's all right. Overall, 2D, and I said 3D, so our wanted level does work. Although, I'm not seeing any actual like uh police logic here, so Next up, should I go this or should I go this way? Go We'll go this way. So, let's try the calculator. We have 84 * 6 504. All right. Good. Next up, we have notes. Okay, simple. But, again, this is a preview model, so we do need to keep that in mind. Wallpaper changing capability, a very nice option of nine specific gradients for us to cycle through. Although it says right click desktop for more options, so that's kind of interesting. Probably stick with what was there first and foremost. Next up our other 3D game, which hopefully is more 3D than the first 3D game. Okay, find the green beacon and we have a timer there. Now, unfortunately I can't move. But it is 3D, which is more than I could say for the GTA. So if we could combine these results, we probably would have had something rather impressive. Now we have a paint feature, which is good. I haven't seen one of these in a while, which is always nice when they have these. And we do have the potential save option, so So right now it's a mouse. I can't draw when the mouse icon is shown. I can only draw when the crosshair is shown, but like sometimes it snaps out of that back into the mouse. This is quite a perplexing paint applic- Oh, you know what it is? I think the window is just like So the dimensions of the actual paint canvas here are what we see here. I'll save that. paint.png Put it in a museum. files Good. Do we have some fake game save data? No, but we have the fake EXE file for the game, so this is an emulation of some form of a Windows operating system. Although none of these are actually like interactable, it could be said. Okay. We do have voice control, so we could do open app, change wallpaper to color, close all, minimize all. Hello says hi back. Hello. >> Hello there. >> That's awesome. And it even showed up in the bottom right like what we had said.
-
7:32
, obre el vídeo en una pestanya nova
Open GTA clone. >> Opening GTA. >> Oh, and it actually gives us um audible response as to what it's doing as well. That's kind of cool. Most of the time when they implement this, I have seen this before as a special feature, it will just perform the action. It won't actually read it back to you, which is more accessibility focused, I suppose could be said. Change background green. >> Command not recognized. >> Yeah, cuz I'm I kind of messed that one up. It did specifically tell us, but oh, we can do a random wallpaper. It seems like this is actually creating gradients that don't exist in that list of wallpaper choices. Ugh, yeah, that lime green one would have had me on a rant still if I'd seen that in the um settings page. So, that's kind of cool and that's why it said in the wallpaper thing, right click desktop for more options. So, it must be procedurally generating random gradients here, which is kind of cool. Overall, this is again a very experimental or preview-like result, but some of the things here were interesting such as the voice accessibility actually speaking back to us when it was performing actions, as well as the random gradient generation. Very interesting. Now, I've made the difficult decision to basically compress the static subway scene test. So, normally we have it generate a beautiful static subway scene, we check it, and then a follow-up prompt is to turn it into an FPS game. I feel like it's just easier to have everything done in one go instead. So, it still has the initial prompt here denoting that it should create this beautiful static subway scene with the brightness slider and ability to move around, but then I've basically appended with the scene done well, turn it into an FPS, and then with all of the traditional things that would be included in the follow-up test for that. Now, something I'm noticing here is this model is very quick, and I'll be interested to see if this like small line of code comparatively to what one would expect persists throughout the test or not. All right, so any question I had about script length has been answered because this was over 2,300 lines of code. All right, so let's try our subway FPS game, and I will check it. Oh. I I thought I heard a thunderstorm, but no. It was just Thunder Compute where you can get the world's cheapest cloud GPUs. Thanks to the homies at Thunder Compute who routinely give me free compute if I ask for it. They also do have a nice new feature implemented where if a GPU is not currently available, you can just sign up to get an email notification when it does
-
9:52
, obre el vídeo en una pestanya nova
become available and then you can go and rent it. All right, Subway Terror Survive the Underground. Toggle flashlight. Okay. First and foremost, I often get tricked by these like oh my goodness. This is significantly better than I expected. I don't know what I expected, but just the sheen of the light appearing and reflecting off the weapon. We even have the flashlight. What was the flashlight button? Uh-oh. I just pressed something. Oh, it was F. It has that like the lag issue, but maybe it will speed up this Wow. Okay, we've got a tube style. I mean, let's just look at this as a scene. Oh, it does have that. I've not seen that since the Opus 4.6 or 4.7 result. Nothing else has ever actually implemented effects of where the ammunition hits in the scenery. This is the second time I've ever seen that. The only other one was a Claude Opus result. Now, I do believe these enemies may actually be backwards because it seems like the face is actually there, but it's just not like let's see what happens if we lose. And the flashlight feature is really cool. This is very very very very well done. Oh, okay. Well, turn the flashlight back on. We have benches. We have trash Did that have a label on it? No, it didn't. Okay. Um they're inverted signs and a lot of the things I'm noticing here are seemingly kind of inverted such as the enemies actually having faces. This was very, very, very cool. I mean, we even have the the shell casings. That that is awesome. Just the This is this is exceptional. It also does have like a a light that's flickering somewhere, too. This is All right. I'm very pleased. Oh. We're going to lose now, but that's all right. And it did have sound effects and things like that. Next up, we're going to do the C++ self-contained Skate Sim result. This is the one with the retro '90s California boardwalk aesthetic. This is a test I find that really seems to separate the big model smell from the big model. So, we'll see how it does. And again, I should mention because I can't actually use this through the API right now, we are just stuck in testing it in the web chat. Otherwise, this is something I very likely would instead run using open code. This is very interesting. So, it's saying the original code uses something Windows-specific, which the prompt does specifically say do not use any external dependencies or make it self-contained with no setup or external assets required. Okay, I guess that could be somewhat vague, but I've never actually had this specific issue. Every other time I've had a result that can be compiled and run on Linux without any
-
12:53
, obre el vídeo en una pestanya nova
extra work. Though, it has proposed an interesting solution here, which is basically to compile it for Windows and then use Wine to run it. I'm actually interested in seeing what happens when we do that. I had to rename the downloaded file, but then hypothetically, okay, we do get a bunch of compilation errors right here, or build errors, which is a little frustrating. Now, the other option it gave us here is to patch the code by deleting everything in the original file from basically here and replace it with the Linux equivalent. So, unfortunately, we were still getting a bunch of compilation errors when just trying to use the Linux script it had given us. So, I gave it the compilation errors, as well as told it to just fix this and give me the full updated script to work on Linux. All right, we have our hypothetically rectified and working for Linux result. Okay, very good so far. So, let's just try it now with skate and we should get a look. So frustrating. And something I've noticed, I was wondering because something I So, I've good to say and I've bad to say. Good to say is the take on the pedestrians here when we are actually facing the direction to see them is quite interesting. It's a unique style that it's chosen to use. There is movement and things of the sort. Now, obviously the visuals are just kind of bunk. I'm trying to see if we can at least get on a rail here and unfortunately we're not able to. C for controls. Okay, so F would be kickflip and air. There we go. So, a lot of the issue right here is just something's not rendering properly. I know folks in the comments are more knowledgeable about the specific issues encountered when it comes to like rendering issues than I. So, please feel free to fill us in on what specifically the issue is here. But I have to say this is a little frustrating and I am I was expecting better and we could see if this was actually like properly rendering things, this could potentially be pretty impressive. The pedestrian density rivals that of what I've seen in some of the new GTA trailers. Uh so, I'm kidding. Um So, and it's frustrating because I don't I don't know how much usage allotment I'll get here, so I don't want to blow it all just on one result. But this is something where if we had this through the API and it was working in open code or something, it would be a fantastic candidate for it to go at this and iteratively improve it. But for now, being that we are stuck using web chat, I am just going to move on to something
-
15:11
, obre el vídeo en una pestanya nova
else. So, I figure why not do a classic, that being the flight combat simulator game where it needs to include three selectable planes, some enemy logic and things of the sort. It will be interesting and because that Subway FPS was just suspiciously fantastic, I am very, very hopeful that we will also get something equally as fantastic here. All right, let's take a look at our flight combat simulator result. Okay, Sky Dominion. Very, very, very, very odd plane image draws here in the selector card, but hopefully those don't translate into the actual game. And they don't. We have very interesting terrain. I don't believe we have sound. Now, again, I didn't specifically tell it it needed to add that, but it's something I would have liked to have seen. We have good flight mechanics, I'm going to say. The actual movement of the plane and things like that. Sometimes that's a big point where models struggle. It's actually quite smooth here and everything of the sort. Now, I'm noticing really the biggest issue is either the ammunition just isn't working or the tracers are not visible. Though, the terrain is very terrain-like with water and the clouds look good. Sometimes shift is boost. Okay, good. That did actually work. As we can see, we have an altimeter as well as a speedometer, just showing us our current knots. The clouds look pretty good. We do have a mini map in the top right. Again, though, I'm noticing no ammo. Let's see what happens if there's actually crash logic. Okay, yeah, there is. There's There definitely is. All right, does our propeller plane have space is to fire? Okay. Does our That's That is good. And can we fire? Yeah, we can. Okay, I actually see the ammo tracers going out from the propeller. It's just really quite difficult to see. Ooh, all right. We do have a lot of hostiles, as he calls them, and we do have Q and E as well for yaw and pitch and things like that. I'm going to say this is a promising result. It's simple, but the flight mechanics are Okay, that was kind of sick when we spiraled and then we saw the ammo tracers. The The mechanics look really good. I don't think the enemies are actually able to hit us. Let's speed up. So, our final plane is the SR-72 Phantom. Yep. Yeah, I could see that. Very um phantom-like color scheme. This thing's inevitably going to be pretty quick. All right, 550 knots. I I also want to just check how it drew the collision boundaries for the ground, if they're actually drawn in the peaks and valleys or if it's just like a flat plane that it drew instead. And it does actually seem like the peaks and valleys
-
17:47
, obre el vídeo en una pestanya nova
are drawn properly with collision. So, what I mean by that is if it just drew a flat plane for the collision detection when flying underneath one of those peaks, we would essentially crash. But, it didn't do that, so this is actually kind of cool. Flying low here's And for good measure. Interesting. And again, because we're stuck using the web chat interface, I figure why not try the historic 3D printer simulation test. I've noticed sometimes some of the Gwen Max models have trouble with some of their spatial like rendering or they'll like render things almost in an inverted position from where they should be. I believe this happened in one of the previous Max tests with the 3D printer sim test. So, it'll be interesting to see how it handles that. As so far, the CSSJS and HTML results that we've received have really been quite a decent, solid, impressive ones. I've no idea what I was just trying to say. They've been good so far. All right, let's take a look at our 3D printer simulation result. Okay, so far so good. I'm seeing a standard Ender 3 style bed slinger. We do have the nozzle, I mean, the filament, but I don't see the nozzle. Oh, okay, it's down there. All right, let's take a peek. Okay. Well, I will say, specially before there has been some pretty severe issues. I'm not noticing that here. We do have just a floating model, but I will say the model floating nozzle, but I will say the model of this nozzle, I don't know why that's difficult to say, but it is, is well done. It has like kind of the block that's actually there as well as that I'm not 100% sure what it is, but it did draw this fairly nicely. I'm going to say if we actually take a look, we definitely can see the layer by layer. Now, if I can slow this down to something that's not like light speed and maybe we'll try that once more. Okay. So, unfortunately it's still like the nozzle's not where it should be and the speed even at 1x is just significantly too high, but I'm going to say the actual shape and things like that works nicely. Let me refresh. We'll try our cylinder at the slowest. Okay, good. The nozzle's actually moving where it should be. It's just like the coordinates are slightly off, but the bed moves well and this is acceptable. It's like a time lapse. It looks like a time lapse from an actual 3D printer. Like this is how the recordings of those look. Okay, and then finally we have the pyramid. Let's see. All right, let's check this one. Oh, wow. A fascinating take on the pyramid
-
20:14
, obre el vídeo en una pestanya nova
with this specific like a It was not as impressive as the Subway game, which is now my bar for judging all the other results in this test. Next up, we're going to be trying a difficult front-end design text. Now, to be completely honest with you, I just had this prompt generated by a different AI because I want something that is going to push this, but basically it needs to do an immersive front-end landing page for Sandglass, which is a fictional creative technology lab. It should showcase a beautiful interactive sand simulation and particle effect as the centerpiece. So, basically we're looking for something very good looking that has a sand particle effect to it as well that works properly. All right, let's take a look at our sand particle effect front-end design. Okay, not necessarily what I wanted and it does have this like I don't know how to describe this effect, but it's one that I've seen before at least in terms of the font and things like this, but it does have an effect here. Okay, I did kind of build a contingency plan as well here, which is going to be Oh, that was actually kind of cool. When we click, watch this. I did have a backup plan here for something completely different that would not go well with the stupid font that seems to be rearing its head into a lot of website generations. All right, so next front-end test, and I am not going to read this all. It's basically for a pasta sauce brand called Ravioli Rosso. And the whole thing is it should have like a The background should look like glossy tomato sauce. The canvas base should have an animated sauce universe as the visual centerpiece. There should be olive oil, basil, steam, and ravioli shapes drifting in a luxurious red-gold liquid world. Clicking should create a satisfying splash burst of tomato droplets, basil flakes, parmesan dust, or tiny ravioli particles. And it should just look premium. Okay, interesting. That was more of thinking than I've seen for some of the other results, so hopefully we'll get something uh good. All right, so here's our pasta sauce high-end website. You know what? This is good. This is fantastic. I truly do believe that going for this result is very fitting based on what it is. This is just gold. Now, there was a chance that like it was going to do all of these with SVGs that it made, which could have ended up being a complete disaster. Okay, the spinning raviolis kind of hilarious. And when you click, it actually makes it look like the sauce is responding. This, like the can of true and things like that. This is fantastic. Viscosity and soul, deeply rich, bright
-
22:41
, obre el vídeo en una pestanya nova
summer bright. Okay, um this will be an ask the audience if anyone notices if that actually made any measurable change to the actual page. Maybe it changed the brightness. Okay, it did. It changed the bright This is This is just like This is peak. You know, it's over for ravioli brand web designers. All right. Three reasons your Sunday gravy just got outperformed. Slow simmered tomatoes. And the cards even have some movement effects on them. This just Look at this. Okay, I could do with that either like make it more centered or just make it like significantly larger out to here. Order a jar, find us in stores. Queens, New York. Yeah, I guess. It could work. This was just like This was very, very good. I'm so I've like so much better than this garbage. Good. All right, next up we're going to do the drum kit simulation where it needs to also include the auto play feature which will play four selectable drum tracks. All right, let's check out our drum kit simulation result. So close to being excellent. If the bass drum was oriented in the correct light way, this would be very good. The speakers are on, so let's just check. Very good. All right. I could turn that up a bit. Phil Collins test was passable. All right. Let's try auto play. Rock groove. Can I play along? Yep. All right. And there's a BPM slider. Okay. So here's funk. Next up, we have jazz swing. All right. Very good. I was looking up at this because I wanted to see if that was going to be played first, and it was. Then finally, we have double time punk. This will be interesting. Okay, overall, not bad. I I'm starting to notice though, it almost feels like whatever this floor plane was, it just wrapped one of the wood materials that would likely have been better around one of the drums. So, that's a little perplexing, but it did do everything. And really, this would have been a stellar model result if the bass drum was oriented in the right way. And I had mentioned previously noticing some odd spatial behavior from some of the Max Quen models. So, this does seem to be an improvement all aside from this. And the effects on some of these individual symbols and things like that were actually kind of cool. I think for the final thing I want to do, because we have not hit any usage limits or anything of the sort, is I want to go back to the self-contained C++ skate game and see if we can get it to a better point. If you recall when we actually ran this, mainly the biggest issue was that a lot of like the
-
26:20
, obre el vídeo en una pestanya nova
surfaces were not properly rendering, and it was just kind of disappointing that being the case. So, I'm going to give it some sparse amount of feedback describing that issue and tell it to generate the full fix script. All right, let's see if our fixed result compiles. It does. So, the next thing to do is just take a peek at it. That is such a significant improvement. I'm very happy that we did this because it went from being like, eh, okay. Although, I will notice all of these pedestrians look like they're being held up. They're just like, so aside from that, this is this is very well done. Let's take a look at some of our tricks. So, F H Okay, H The riders is very slow at 3 miles an hour. As a matter of fact, the pedestrians, we do even have some foot movement here. Look at the water effects, the palm trees, everything there looks good. Can we grind on this bench? Nope, I'm just trying to get to one of these rails. So, unfortunately, I don't think we can grind. One, two, and three are just spin. All right. Wow. This is just a significant, significant improvement over what we'd previously seen. So, I wanted to to test that at least. It's still, you know, could use some work, but from where we started to where we ended up, much, much better. So, that's going to conclude our first look and test of the Quake 3.7 Max preview. I do just want to make sure like, so I'm just going to check one final time cuz sometimes, but rarely, that's happened. Okay, so no, I still don't see any official proper announcement posts or anything like that, but I would go out on a limb here and definitely say that based off of what we saw from some of these results, this is an improvement. Now, the browser OS was surprisingly, or maybe not surprisingly, the least impressive result, I think, out of everything we saw. Hands down, what really blew me away the most, well, would be that pasta sauce website, very likely, but a side from that, the thing I really liked was the Subway game. I mean, okay, it's going to be too loud here, but this was just phenomenal. And as a matter of fact, the only other model that I've ever seen that actually has ammunition hits reflect in the environment itself like that was Opus 4.6 or Opus 4.7. I don't 100% remember. So, this was just phenomenal. And we even had the flashlight toggle, like flashlight on and off, which was just This was really, really well done. So, that was definitely the most impressive result I think for me personally. Well, I mean again, this was just artistic
-
28:56
, obre el vídeo en una pestanya nova
genius, but aside from that, we had our flight simulator game. The propeller plane was good. The flight logic worked well. It did properly have collision logic and things like this, but this was nothing like super mind-blowing comparatively, but it was a solid decent result. 3D printer sim is the same. This is an improvement at least over some of the previous ones I've seen, but overall still kind of buggy, although not to a huge degree. We also did have the drum kit, which kind of had some similar oddities to it as the 3D printer sim just in terms of the orientation that some things were drawn in the actual space. We had our first front end test, which was the sand glass one, which I'm just not a fan of I guess this color scheme, but fortunately the Ravioli Rosso website was a complete just winner for me. I I don't know why. You may be like, "Why does he like this?" And I don't believe that anything on Earth could properly answer that. So, make of that what you will, but I personally just found this to be fantastic. I don't know. It's just the absurdity of the result, but um learned to love it cuz this is a test that's very likely going to be sticking around now as an additional front end test, which I'm quite excited about. In terms of other things, of course we did have the C++ skate game, which became a significantly improved result over what we had previously had, and I feel like there was something else, but I do not 100% recall. No. Okay. Well, that is going to conclude then our results overview and our first look at Quake 3.7 Max preview. Now, you may notice that in addition to that, there is also a Quake 3.7 plus preview. I may or may not do a dedicated video on that in the same vein as this. However, because Max is one that is always historically and likely will continue to stay closed source, I wanted to test this, but it is very possible that the Quen 3.7 plus model may have an open source analogous version to it which would be really cool to see. So, I want to wait a little till we learn some more about the Quen 3.7 family to do that. So, that's going to wrap up our video. If you have any questions, please feel free to leave them in the comments or on the Discord and thanks for watching.