Startracker 🔺

9.5K posts

Startracker 🔺 banner
Startracker 🔺

Startracker 🔺

@startracker

Vibe coding, creative technology, applied AI. Building agentic pipelines for creative, video and online game production.

Tokyo Katılım Ekim 2012
1.8K Takip Edilen6.1K Takipçiler
Sabitlenmiş Tweet
Startracker 🔺
Startracker 🔺@startracker·
Day 11 of building the space game I saw in a dream. ⛈️🚀 GPT 5.6 Sol's [SKY] Storm team cooked this time, but only as phase 1 implementation. Let's take a look: ☀️Stood on the runway and filmed some 2-3 minutes of overcast rolling in and a storm forming over the island. no scripts, no cutscenes, that's the live weather director shifting states in real time, sampling the same wind field the aircraft and every blade of grass actually feel. This sky is the stand-in. it passed its screen test today, so now it earns the full rebuild — reference-matched cumulus with flat bases and sculpted tops, rain you can see AND hear soak the apron, lightning that glows from inside the cloud. I personally like how some of the shots turned out, and bit by bit, I'm getting enough elements to build out the entire island. I've seen a lot of people tackling clouds too, so I hope with everyone at it, we'll soon get a new vibed cloud industry standard! Clouds are important! Not only on Earth, they're the foundation for Venus-like planets as well as gas giants or even exotic objects.
English
6
1
19
1K
Startracker 🔺
Startracker 🔺@startracker·
@ziwenxu_ I think the ceiling is very high with it although agent iteration speed is absolute snail and a token guzzler. But I'm enjoying it!
English
1
0
1
13
Ziwen
Ziwen@ziwenxu_·
@startracker Let’s goo!! Feel satisfied see another person building on unreal
English
1
0
2
50
Startracker 🔺
Startracker 🔺@startracker·
Day 11 of building the space game I saw in a dream. ⛈️🚀 GPT 5.6 Sol's [SKY] Storm team cooked this time, but only as phase 1 implementation. Let's take a look: ☀️Stood on the runway and filmed some 2-3 minutes of overcast rolling in and a storm forming over the island. no scripts, no cutscenes, that's the live weather director shifting states in real time, sampling the same wind field the aircraft and every blade of grass actually feel. This sky is the stand-in. it passed its screen test today, so now it earns the full rebuild — reference-matched cumulus with flat bases and sculpted tops, rain you can see AND hear soak the apron, lightning that glows from inside the cloud. I personally like how some of the shots turned out, and bit by bit, I'm getting enough elements to build out the entire island. I've seen a lot of people tackling clouds too, so I hope with everyone at it, we'll soon get a new vibed cloud industry standard! Clouds are important! Not only on Earth, they're the foundation for Venus-like planets as well as gas giants or even exotic objects.
English
6
1
19
1K
Startracker 🔺
Startracker 🔺@startracker·
@mattshumer_ @danveloper The prompt works - and I used it but it doesn't make a polished game. And balancing / fixing broken systems afterwards takes time. I use it when I wanna build a new map to get me started but then it's a lot of work to improve it
English
0
0
0
12
Startracker 🔺
Startracker 🔺@startracker·
@martin_casado Yeah similar experience. It sometimes also simply stops working while claiming it's still working. And even lies from time to time.
English
0
0
0
8
martin_casado
martin_casado@martin_casado·
@startracker Yeah it's good at 3D. But really spiky on general systems work. So I'm constantly having to double check.
English
1
0
1
78
martin_casado
martin_casado@martin_casado·
I still love Opus 4.8. And Fable for serious debugging. I find Opus 5.0 almost unusable.
English
58
3
186
11.6K
Fer
Fer@harderthanfire·
@startracker The fast model toggle is still there in codex...
English
1
0
1
56
Startracker 🔺
Startracker 🔺@startracker·
This is actually big - traditionally it was 1.5x speed for 2x the cost, but a higher increase in speed compared to cost. Game theory says we should always run Fast mode - since the pricing incentivizes it. Normal speed (1.0x) = 1.0x price Fast speed (2.5x) = 2.0x price There's absolutely no reason to run normal mode at this point. The caveat being *UP TO* which means it won't always be 2.5x but I believe it's still gonna be a significant discount. Problem is - for now, only for API (but it means Cursor and others should have it as an option).
OpenAI@OpenAI

Along with the price reduction on GPT-5.6 Luna and Terra, Fast mode for GPT-5.6 Sol in the API delivers up to 2.5x the speed of Standard processing at 2x the Standard price. Fast mode gives API customers faster access to GPT-5.6 Sol, with no change in intelligence.

English
3
0
13
1.3K
Jules
Jules@juleshatton·
@startracker I think fast mode will be an API-only option for a long time, sadly.
English
1
0
2
47
Tibo
Tibo@thsottiaux·
We kept ourselves busy with GPT-5.6 Sol. It's pretty good at things. Rolling out - 80% lower prices on Luna - 20% lower prices on Terra - faster speeds on GPT-5.6 Sol in /fast mode - ~ 10X cheaper auto-approve mode ("review for me" in the app), which now uses Luna to prevent many high risk actions from the main agent Towards intelligence too cheap to meter and many more breakthroughs ahead.
OpenAI@OpenAI

We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API. Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

English
795
245
6.1K
454.9K
Startracker 🔺
Startracker 🔺@startracker·
@sama Hopefully coming to Codex as well we have the need for speed!
English
0
0
0
182
Sam Altman
Sam Altman@sama·
major price cuts today: *80% drop for GPT-5.6 Luna, now $0.20 per million input tokens and $1.20 per million output *20% drop for GPT-5.6 Terra, to $2/$12 *GPT-5.6 Sol gets Fast mode in the API, up to 2.5x the speed for 2x the price, same intelligence
Sam Altman tweet media
English
881
669
11.7K
1.1M
Startracker 🔺
Startracker 🔺@startracker·
Grok 4.5 is actually a solid workhorse, as Elon puts it. I use it daily in Cursor, it's always on some promotions there so it's free! Basically it's my favorite subagent, it's fast, and gets stuff done when instructed well! Besides Grok build is built by the same superteam that built Cursor so they know ball.
English
2
0
4
70
robot 2.0
robot 2.0@alightinastorm·
used grok build for a full day (non gaming) i.. have to admit.. quite pleasant experience? it's fast it's smart it's reliable is grok actually back?
English
19
0
47
2K
Startracker 🔺
Startracker 🔺@startracker·
@yang_frog Lately I'm very unhappy with Opus, it feels really nerfed. At least the one that comes with Claude Code. It does decently when ran from Cursor. So I mostly use GPT 5.6, Fable or K3 for big stuff and Grok as free labor!
English
1
0
2
49
Startracker 🔺
Startracker 🔺@startracker·
@CuiMao I didn't expect CuiMao to make a CG donghua anime, but here we are, and I hope there's more to come! This is awesome!
English
0
0
0
147
CuiMao
CuiMao@CuiMao·
大家好,我们即将迎来Seedance2.5,请允许我榨干Seedance2.0所有性能配合最新Kimi K3模型为这个人类历史上具有里程碑意义的视频模型画上一个完美的句号。【完整游戏流程请移步至评论区点击链接体验】
中文
243
127
1K
188.9K
Token Gremlin
Token Gremlin@TokenGremlin·
Do you guys remember when I said I suspected the model launching in August wouldn’t be the same one from the Hugging Face incident? Yeah, I was right. But now I have no idea whether the model being released will be stronger or weaker than that one. I hope it comes out as GPT-6 and is even better, but the odds are pretty low.
Token Gremlin@TokenGremlin

@developed_life I'd say there's about an 80% chance. The only thing I'm still unsure about is whether the specific model I mentioned will actually be the one that gets released, or if they'll ship a weaker checkpoint under a different name instead.

English
2
0
10
1.2K
Jeremy Nguyen ✍🏼 🚢
Jeremy Nguyen ✍🏼 🚢@JeremyNguyenPhD·
Is it just me or can Opus 5 be quite unpleasant to talk to if you get off on the wrong foot? Keeps attributing motivations to my questions that are not there. Or acting like I'm arguing when I'm being curious. I've always previously found Claude and Opus to be the ones I liked talking to the best.
English
40
3
229
12.4K
Quasar J
Quasar J@Quasar_J_Music·
@startracker Would you be willing to look at something I'm working on and give any advice?
English
1
0
1
12
Startracker 🔺
Startracker 🔺@startracker·
Day 10 of building the space game I saw in a dream. 🌙🚀 FIRST BIOME results - brought to you by OPUS 5! Fable 5 did the research, the geology and the math. Grok 4.5 did information gathering and retrieval. 2x Opus 5 agents crafted the initial procedural world. ☀️God rays through the trees. Grass lit by real time global illumination, every blade catching the sun no baked lighting anywhere, this is the same sun my physics engine tracks for time of day, computed live. This little island taught me more than any system I shipped, including my hardest lesson: never fix a small problem with a bigger one. This biome will not make it into the main game. Unfortunately there was a lot of issues with how it sees tectonic movements and elevation rules, so the team is already working on another iteration that will hopefully work. Still, I wanted to show you something that looks lovely! Because I simply enjoy running around this island's meadows. Meanwhile, behind the scenes: the tow tug is getting true two-body physics, the storm system is being rebuilt against reference photography, and today I ordered our biggest program yet -a planetary world generator whose terrain has to statistically match real geological survey data of Iceland before I let it grow other planets. Enjoy the video. AAA yet? 📽️
English
11
2
66
4.4K
Startracker 🔺
Startracker 🔺@startracker·
This is not a one-shot prompt. It's already 10 days with billions of tokens in. One-shot prompts are demos for engagement. If you want to build something like this, it will take time, a lot of effort, and constantly juggling many agents. I suggest you start by organizing your project into several teams - one working on geology, one working on biology, one working on physics, and then 3D art team. (this is what I do) And then have a CTO to manage them all, this agent should know your vision the most and will write precise technically specified prompts for each time how to handle things. Just go ahead and try it. You'll have some results very quickly to get you motivated, but then it's up to you to chase top quality
English
2
0
1
25
Quasar J
Quasar J@Quasar_J_Music·
How??? How do these look so awesome? I'm trying to build a game myself, albeit a very different style but dang I'm slightly depressed trying to figure out how ya'll are doing this so well. I do understand the one shot prompts and such but as a non-coder trying my best doesn't appear like it's going to be good enough.. oi.
English
1
0
1
41
Startracker 🔺
Startracker 🔺@startracker·
@mikeluan123 This is pretty cool! I love the Homeworld series and some of the motifs always follow me even into the one I'm working on rn. I wonder how far can this be taken if done over a few weeks.
English
1
0
0
56
nooooob
nooooob@mikeluan123·
It's playable now. The Opus 5 RTS from yesterday, running in your browser. 13 ship classes, procedural hulls, planets, effects. No image files anywhere in the build. Every pixel generated at load. $632.65 of Opus 5. MIT. e01.ai/starfall RTS(RealTimeSlop),technically.
English
8
8
224
8.8K
Token Gremlin
Token Gremlin@TokenGremlin·
I tried recreating the test from my previous post with GPT-5.6 Sol, and the result was also very good. That said, Opus still feels clearly ahead in game design and 3D work right now, especially in visual composition, environment detail, material quality, and overall scene coherence. I am very curious to see whether GPT-6 can surpass both. At the moment, though, Anthropic seems to have a real lead in this category.
English
4
4
28
1.7K