Does anyone have any evidence that having the most powerful model being the orchestrator, and other sub agents being progressively less powerful, is optimal? Is this just common sense or it has been measured somehow?
For me it could be that this scheme may make the code itself poorly written, verbose or faulty
I think whether it's bottom-up or top-down, the core principle is still that you use cheap models for grunt work and use frontier models for expertise and direction.
In my setup, the orchestrator also drives a lot of the design and planning, and the less capable models get inputs from the more capable models by responding back up the chain. The more capable models also review everything that is produced by the cheaper models.
I've also had systems that invert this and allow cheaper models to call out to more capable models when they need to, but in my observation I found them quite bad at knowing even when to call out and I find wrapping the cheap models a bit more reliable in that problems tend to get caught before they bubble up to the user.
It sure saves tokens and even the 100$ subscriptions provide massive amounts of dev power. Also fair to note - mixing models is perhaps a good idea anyway as this way the entropy in them mixes and this supposedly improves result quality.
https://www.nature.com/articles/s42256-026-01268-y/ This is probably what you're looking for. TLDR; weaker orchestrator + strong workers scored higher than the reverse. Caveat with that, old paper: "High-capability models: GPT-5, Claude Sonnet 4.5, Gemini-2.5 Pro; low-capability: GPT-5 nano, Claude Sonnet 3.7, Gemini-2.0 Flash."
With that said, there is evidence that we recently crossed over a line for 'best practices' when it comes to systems of multiple agents. Comparing the shapes of 'n agents working collaboratively' vs '1 agent spawning n agents to handle distinct work', Opus 5.5[1] and Opus 5[4] performs better with the former, Sonnet 5.5[2] and Opus 4.8[3] performs better with the ladder.
Thanks for sharing this! I didn't see GPT 6.1 Sol on OpenCode Go, are you using that through some other service? For Muse Spark 1.3 Contributor, I see on openrouter that it is not ZDR. Given Meta's track record of privacy abuses, are you concerned at all about that?
The unfortunate reality is that all of the fun recompilation projects are helping seal the fate of us not being able to run games, or software in general, on our own hardware. The risk is very quickly becoming too great for proprietary software publishers in letting compiled artifacts land in customers' hands.
I feel GTA 6 will be the last mainline GTA that we run on our local systems.
That’s actually a really interesting take I hadn’t thought of. I’m involved in a decomp project (that to my knowledge is all human created) and knew there would be a legal reconning for AI powered projects, but I didn’t anticipate the business model changing to prevent near future releases from being preserved through fair use RE. What a bleak outlook.
This is not necessarily the case, as you still need to deliver an experience that customers enjoy and will pay money for. This means consoles that let them play games in their hands with the highest picture quality and lowest latency. No disruptions from poor wireless or significant interference in environments like apartment complexes.
Fully homomorphic encrypted processing may also become a norm and if that is combined with some games going back to being console only, there could be significant enough protection for games that contain trade secrets. That said, the days may be numbered for trade secrets at all given that better models may make today's clever trade secrets look juvenile.
At the same time, a lot of these games were already getting decompiled and inspected at runtime. That has historically been one part of the incentive to release games on console only or console first, particularly first party titles.
In any market where it's relevant to their business interests, they can still rely on the copyright laws and can set traps for any companies using decompiled code as part of their own paid and sold product. That whole process will become cheaper and easier to do as well, plus new approaches to slowing down or hindering the results of decompilation will become normal for games in mathematically provable ways. We've already seen approaches over the years like Denuvo.
Always online games themselves have benefited from some of their code being in the cloud and that will obviously be under fire more than ever with people writing their own servers, but even that might have novel approaches that can survive advanced tampering.
Indie platformers will continue to run entirely on the console. But AAA games have a history of prioritizing visuals over everything else, because that sells trailers. Fully streamed games executing on a remote GPU allow better graphics and prevent decompilation and piracy, all great things from the publisher's POV. So far the economics haven't been appealing and the experience isn't uniform for customers, but I could see AI-assisted decompilation pushing the AAA publishers over the threshold
If you can't profit off of your product sufficiently, because not enough people are interested in the experience that streaming provides, then you can no longer justify even making those games and it all becomes irrelevant. So, in the end you release your game to local hardware anyway.
They can't simply copy your game and reship it, because that would be against the copyright laws in most places where these companies make their profits from. Making something sufficiently new would be technically legal, but it would also take massive computing resources even with LLMs. That's either local compute time or cloud spend.
I think the bigger risk is just brand new games being generated of all kinds rather than the more expensive and legally risky full decompilations. Mass generation could start saturating the market and soaking attention. The low end indie market is already pretty saturated and in the end games have to be appealing, review sufficiently well and find an audience. Plenty of good games have barely sold any copies.
So almost a bigger question will be, how will AI impact the marketing industry?
Games can now be recompiled to other platforms, like consoles and the browser, via WASM. So Zelda BotW can be played on a jailbroken PS5, and in the browser.
What actually limits someone from doing this?
1) You need the hardware to play the games, still. You still need a good enough PC or a console like a PS5 to play all these emulated/cross-compiled games. Though, the cost is moderate these days: a used PS5 goes for $400-500?
2) You need to want to jailbreak your console, giving up "mainstream" features (multiplayer, warranty). This requires some interest and know-how, too. Though the jailbreak steps are easy these days.
3) If you lack a console, your PC needs to be strong enough to play WASM or emulated/cross-compiled games. I suppose for consoles like the Switch, that aren't super powerful, most PCs made the past few years could play a AAA Zelda game in the browser. But playing on PC is often not as fun as on a console, with the couch and controller and all.
I have a hard time people en masse want to give up multiplayer, by jailbreaking their consoles.
But single-player games are probably going to have a harder time.
The video game world is already super top heavy, with mass-market multi-player games (like Valorant), not to mention casino-like gacha games. So the trends may further favor this current track.
Sooo we can conclude most of these trumped up recompilation projects are a coordinated attack, so decision makers could say, we need to protecc ourselves from this thread.
Someone started it, lots of other bored people, followed. Damage done, remedy already in store for gamers.
I'm not worried at all. LLMs are already so good that they can reproduce many apps and games merely by using those apps and games for a little while. By the time customers are willing to accept streaming only games (which I contend is still MANY years away from mainstream), LLMs will be able to near perfectly replicate these games. Or better yet, make far better versions of them. I'm not even sure we have a games industry in a few years.
BAR has quietly become the best RTS to release in years and is a huge amount of fun. It really feels like the version of TA that my nostalgia remembers even though the game and units are a bit different.
My group don't play competitively, we hop on for a bot bash once a week and the AI is a lot of fun to play against. I'm excited for them to finally release on Steam because that will lower the barrier to entry and increase exposure because the game definitely deserves success.
The maanHimself repository appears to be 35 hours older than the aloshdenny one based on `created_at` from the GitHub API. maanHimself's `pushed_at` predates aloshdenny's `created_at` too.
That doesn't guarantee maanHimself is the original author, but it's looking likely.
There's an issue with GPT-5.6 Sol where it sometimes starts mixing thinking with output and stops working[1]. Once it starts doing that, the session is essentially cooked and you need to do a bit of gymnastics if you want to recover it.
This happens to me more commonly in large projects (>100k LOC) and in those projects it seems to happen every few sessions. I feel this specific benchmark would be impacted by this more than the smaller contrived benchmarks.
I can't help but imagine agents using caveman speak sometimes start behaving in a stereotypically caveman manner, even if it's subtle. Is there a chance the agent does less reasoning because of it?
Caveman invented fire, the wheel, domesticated wild plants and animals, organised society, survived the Toba catastrophe, cooked food, and was having sex ages before you and me. Don't write him off as stupid.
And I wonder how they actually spoke. Since there was no visual communications medium except for cave art. (Some of which is very excellent. Try drawing 3d curved horns in perspective.) So people would have used verbal communication more. Also no written word. So one would expect there to be quite a lot of oral tradition. Like people reciting poem form epics.
If we assume the time is before farming, population density would have been low and limiting culture. Hunter-gatherers might have travelled a lot more than farmers with a homestead though.
The Pleiades cluster is called the seven sisters in Greek. That's curious, because the human eye under the best conditions can discern only six stars in there. Even more curious, the aboriginal Australians also called this cluster the seven sisters.
Ancient Greeks' and aboriginal Australians' last common ancestors split about 60,000 years ago. And astronomers tell us that 60,000 years ago, there were seven discernable stars in that cluster.
One could this conclude not only is speech likely 60,000 years old, but also that the tale of the seven sisters might be a tale from so long ago.
Interestingly, back in my ill-spent youth, a few of my fellow astronomers and I were out in a very, very dark-sky location in the late 70s/early 80s, and we were able to consistently count and draw between 9 and 11 stars. Although we would tease those who could see 11 stars as using averted imagination. :-) Today, if I can see six stars, it's an okay night in an okay sky.
fwiw, if you can get out to dark skies where you can see fifth- or sixth-magnitude stars with the naked eye, I highly recommend getting out there when it's a low-moisture atmosphere and the Milky Way through Cassiopeia and Perseus is vertical, as it's a rather dramatic sight of this stream of stars heading down to the northern horizon.
The summertime Milky Way overhead down to Sagittarius tends to get all the love, but the wintertime Milky Way is also visually rich and worth spending time on.
It's not exactly a joke, it does reduce the amount of tokens. However, it does not improve performance (fine tunes are finnecky things, hard to get one right).
Personally the only 'enthusiast' modified qwen 3.6 27b or 3.6 35b-a3b I've found useful are the ones that have been run through heretic and adversarial data sets for innocent/dangerous prompts, to produce uncensored LLMs. They have some niche non-coding uses for things that a commercial LLM will never talk about.
I think those are mostly vapor that runs on the small culture of "models should not be censored" thing. But from my experience, they unlock nothing meaningful.
Fine-tuning is great for really small models on specific applications, but it's not something that can essentially improve a more generic model.
That said, there seems to be a fine line in quantization+finetuning that could recover performance. It's just hard to get a hold of it (I feel it in some models, but it's hard to say yet; lots of small labs working on this RN).
The most interesting use I've found for them so far is strictly as a novelty. Give a chat session with one to a completely non technical person, who at least knows that openai and anthropic have some guard rails on stuff, and tell them to wild with something like "give me the precursors and chemical formulas for the precusors for crystal meth" and watch it answer.
But does it answer those queries correctly, or does it just not refuse to not halucinate an incorrect answer? From where would it even have that information?
I don't know enough chemistry to say one way or the other if it's just wildly hallucinating the precursors and processes, but it'll also do things like, write an ISIS press release, or similar. There's a data set of basically a bunch of antisocial or dangerous prompts that some people have got variants of qwen to pass with 0 out of 465 refusals:
Just so we are clear, no "caveman" spoke English. "Caveman speak" is just shortening the vocabulary of english, not a "caveman language". Given this, your concerns for "stereotypical caveman manner" makes very little sense since what caveman are you talking about?
The concern is not that the model was trained on actual caveman artifacts, rather on modern media representations of the stereotypical caveman (that never actually existed).
For the vast majority of models I've worked with, 150k tokens of context seems like some magic spot. If you can stay under it then the session will stay on track and the agent won't start getting forgetful and weird.
Just like with real work, it's valuable to break down big tasks into small tasks that you can knock over in a single session. When a session does start getting too big, you just need to ask the agent to reply with a comprehensive handover report and paste it into a new session.
For my quirky workflows I need 100K even to prep Opus 4.8 or higher to begin good work. Not programming, but a complex synthetic architecture for implementing more attentional autonomy for machines.
For those trying to get it to work in OpenCode with a Qwen Cloud Token Plan, this is what worked for me. Note that I've just matched Qwen 3.7 Max for the limits as I don't know exactly what they are.
also, be very careful which API endpoint and API Token you use. make sure you use right one (obseve your quota is used up. if you hit right endpoint quota used almost immediately). so that you do not accidentally burn API endpoint tokens (they are expensive, can easily hit 200 USD / 3 days which do not count towards your membership "Credits", if you say purchased it with 200 USD signup bonus in Alibaba Cloud)
Possibly not the best example but I do feel like Alley Cat looked great in CGA: https://i.ibb.co/wNDvgzQ8/alleycat.gif
reply