If DeepSeek was smart enough. There's a huge risk that it isn't since the only model that ultimately succeeded was Opus 5.5 (sol and astra were tried before).
It might, but also might not. I would imagine the economics are genuinely very tangled up even before you start obfuscating it - it’s very hard to fairly allocate fixed costs between training and inference.
I think a clearer picture would be from inference-only orgs.
TypeScript team member here - this is impressive work, and it's honestly crazy that 3 of these ports have popped up in the last week! I'm currently AFK but we're hoping to learn more and we'll have more to say on this soon.
I think the future of this is -- get your AI agent to investigate this port for anything actually worthwhile, comparing it to the go version, and then take those good ideas and actually maybe bring them into the real compiler.
I'm curious how much unsafe this uses? Are these ports making idiomatic code or just using unsafe all over. There doesn't seem to be anything in the readme.
People were doing silly things with expensive MacBooks all along. It is akin to a non-techie parent saying "what are you doing with your expensive computer ..."
I have to wonder if LLM reimplementations verified against years of human tests will have holes where original authors thought that tests are not necessary and common sense is enough.
I have to wonder this to preserve my ego as a human. I wonder have to this because I sure as hell am never going to go through all that code to check.
Of course they will. For about 3 days, and then it will get fixed.
At this point, for software which has an extensive test suite (or a large body of interoperable software), humans are more likely to create these kinds of gaps than the LLMs are.
As much as possible mechanical (non-LLM) verification of equivalence that generates good diagnostic messages, mechanical translation steps that generate good diagnostic messages, a good memory system to reduce rework...
It is beginning to look like the future is treating Rust as an optimization pass. Write your program in a higher-level language like Typescript, Go, Python, etc. and then have LLMs compile that to Rust.
Not really, the future are natural languages, with some markdown files for formal specification, yaml or whatever, and then Assembly gets generated directly.
3 GLs are going to slowly fade out as LLM tooling improves, and coding looks increasingly as envisioned by 4 GLs advocates.
Sounds awful, natural language sucks for describing complex problems and relationships. That's why people have historically reached for diagrams and pseudo code.
Given the rate of progress, it's hard to argue why this couldn't be a reality.
If in the future Taalas-type chips with baked in frontier models are commonplace, this should be quite different price-wise from what we have now.
It's a bit dizzying to imagine that LLMs might even be able operate at speeds of chatjimmy.ai, but we already have working tech that does exactly this (albeit with less intelligence/capabilities).
Theoretically the LLMs could go straight to machine code. No need for any intermediate language. However, the HN stories have made it clear that LLMs are not great at going directly to machine code or Rust. Rust is, time and time again, only being used as an optimization intermediary. If LLMs were better to target it directly they wouldn't be always be stories about translating to it from another language.
Yesterday I asked Claude to make a RISCV emulator like qemu in x86 assembler. It runs about 5x slower than qemu. That's a pure interpreter vs one of the best JITs. Qemu can run many more instructions though.
We have to accept that those of us that survive the AI layoff wave and want to work in IT, will have to work as technical architects, solution analysts, requirements management, or move into management roles.
Coding as we knew it will be left to weekends and retro-computing handcraft fairs.
Maybe there are just so many of these but at this point for vibe coded stuff like this I’m just like “who cares.” Cool LLMs can make stuff like this now. Unless this reaches some critical mass of usage among developers why would I use it. The more important question is “what does this teach us?” And idk the answer to that one.
Generally building something is not about "What does this teach us?" but rather "What can this be used for?". You don't build a house and as "What does this teach us?". It's honestly almost a perverse question to primarily ask for something that aims to have practical value.
The clear reason this has worth is that it's much faster than TSC.
HN is going to be a weird place to be between people loving AI and the loss of loving tech. I don’t remember last time I watched a good tech dev video. At least the video game industry is a little bit shielded from this but soon it will be the same when we see the speed AI is advancing.
I don't see at all that this is stifling love of tech. Sure if you liked to engage in language wars or vim vs emacs or tabs vs spaces then yes that's over. But tech as we know it is currently exploding, a ton of stuff that simply was not practically possible before is now possible.
Give us some examples of things that were not practically possible. Things are practically possible if the time or money is wisely spent, even if it takes a human a bit longer.
Money and time are precisely the point. Things are not practically possible of a X would cost 100K to develop and it's worth was 30K. But now that it costs 2K to develop it is suddenly practically possible.
2. The abundance of game decomps and recomps becoming available. Literally dozens of example in this space.
3. Reverse engineering of obscure software. A ton of oppertunities lie here.
4. As this posts indicate, severely accelerating existing software by re-writing it in language that are not terrible for performance.
5. Analyzing large volumes of unstructured information.
6. Creating one-off engineering and scientific tools.
7. Beating essentially all traditional Linux distribution, see Omarchy.
8. A personal example: I'm remaking the engine of an old game commercial game in rust.
I'm honestly a bit taken aback that this question is even asked. The volume of software is exploding and the power of the individual has increased a hundredfold.
I don’t quite follow. If you bring software worth into question, as you pointed out everyone’s vibe coding each of your examples so their individual worth has dropped significantly (with the unknown stability of support and long term plans).
My thought was LLMs accelerate the best and worst of development, but what couldn’t be done (outside of a large lab or company) by regular humans with an infinite time scale?
I think the difference here is that the original motivation for choosing Go over Rust was that Go was easier to port TypeScript to.
LLMs change the calculus a lot, the result seems faster, and it's possible that the TypeScript team might want to change directions. It'd be a big deal, but I bet they'll at least discuss it.
Given the Typescript compiler rewrites that keep popping up, Next.js has one, this is another one, Angular just announced they are also going with theirs, eventually there will some community pressure on how the team will proceed.
Note there is already a feedback post from them (currently top one), that they will need to react to this.
Do we know why? Is there some hole the official one isn’t filling? Seems like having 10 different compiler implementations would be a portability nightmare. Now I can’t guarantee my typescript is going to run the same on node/angular/whatever
Yes. This has nothing to do with people trying to rewrite things in Rust for fun. It's a part of his development project that I'm sure will get a video in the future.
Would gold be valuable if anyone could create it at will? Sure, a few people would like the colour and the way it shines, but most would be indifferent. A lot of these things were cool because we knew they were difficult and had technical challenges to overcome. Now it's easy, I just don't care.
In this case the idea here is ‘rewrite tsc in Rust” not exactly earth shattering. If somebody hand wrote that, then they could have gleaned some new insights about Rust or Typescript or something which could be interesting, but in this case they just had an LLM do it and didn’t even read the code. So it’s a big who cares. If it gets adopted by the official compiler team, then I’ll be interested. But right now you would be an idiot to use this for anything mission critical.
When people post AI projects with genuinely new/interesting ideas, then I am interested.
LLMs have been trained to fix compile errors in a loop. Give them years of human labor worth of tests and they will fix compile errors until it works (as well as the tests can ensure). Now you have a codebase no one has read. Good luck adding new code to it.
Let's also not forget that Bun was claiming a rewrite in 11 days or whatever it was, but actually spent months of human labor fixing all of the issues their rewrite introduced before actually shipping it as a release.
Interestingly, your whole comment implicitly has the answer to this. In the future you don't add code or read it, you only add test cases and have the loops work on covering the new case.
I don't know that I agree, especially with LLMs as they are right now.
Maintainability isn't just for humans. An LLM working in an unmaintainable codebase has the same problems we do. It doesn't necessarily follow that if your code can pass your tests today then it will be able to pass the tests of tomorrow. I'm in healthcare. Regulations and payor rules change constantly. I don't know what tomorrow will bring, but I do know that the code I have an LLM generate tends to focus on the here and now (at the expense of the future). Sometimes that's okay, though. I don't know much about writing a compiler, so it could theoretically work.
This is by no means an anti-AI post (I use the hell out of these tools). But I can still see the cracks in them.
The existing test cases were created based on understanding the code. Creating good tests that cover every case you care about without understanding the code is not possible.
These big ports we are hearing about are of projects with hundreds (prob thousands) of unit tests. Not only are the models typically trained with the original code bases in their dataset, but the test suites themselves have project architecture encoded in them..
This was a lot of tokens to spend on a port. I wonder if all of the porting activity out there would be better supported by investing in really good deterministic automatic translators that do 90% of the job cheaply and delegate any decisions that have to be made back out to the AI agent?
What a waste of power... $424.000 of API priced tokens amounts to how many kWh burned? And how would you trust this to build your code?
Motivations says it all:
Motivations
Test model capabilities
Make a fast TypeScript type checker
Make a ts checker that can work in WASM with high performance
Memes
Do you read the test suite of all the programs you run? I trust software because the authors have shown themselves to be competent and trustworthy. I expect authors to understand what they are doing, not just bashing into the guardrails until they get something that passes all the tests.
An LLM has reimplemented a 2 times faster version of a program developed by a team of dozens of brilliant engineers led by few of the biggest experts in compilers and programming languages in the world.
While maintaining 100% test coverage and so far compiling every single project that the original compiler did.
If your takeaway here is "yeah, but the other one has been checked by dozens of paid devs and an entire community", and still those dozens of devs with the backing of thousands of contributors could not make a compiler faster I can't buy say: "maybe we humans aren't as good as writing code as we thought and LLMs are exposing us".
There are companies that spend the equivalent of that in a single day on tokens. Lets not wage war against one person here.
Its the equivalent of telling individual people they are the problem when it comes to leaving their living room lights on when not in the room or they don't recycle all their stuff correctly when companies/corporations have orders of magnitude more waste.
>Spent over $400,000 in API priced tokens with GPT-5.6 Sol and GPT 6 Astra. They wrote over 1.3m lines of Rust over multiple months of /goal loops and never got past like 84% compat.
Then they let Claude code lose on the problem:
> I figured it'd be fun to throw Opus 5.5 at this. It had a working v0 in 10 hours. I assumed it kept using the code the Codex models wrote. I was wrong. Opus 5.5 started from scratch. It got further than Astra in 1/10th the time.
In the warnings from README.md
> Also worth mentioning: I've never read a line of this code.
Looks like this is a company’s org which heavily relies on AI. Probably “the cost of doing business” and to see just what can be done with the new model.
He didn’t spend that much either. That’s the equivalent API cost using his $200/month subscription. It took less than a day so I guess it cost about $7 in actual money.
Ok, that makes more sense. If the other comments are correct in saying this was done using a $200/mo personal subscription, I get this is more of a “what is possible, and how much would it cost without subsidized pricing” type of experiment. Thank you for clarifying on that.
What the hell is "theoretical money"? It sounds like either OpenAI took the $400k hit, or their API pricing is utterly bonkers and therefore completely pointless to quote. It would be like me saying my car is worth $400k in theoretical money because, yeah, I'd totally sell it to you for $400k.
I've been rinsing Opus 5.5 and I've been unable to get over 20% of my plan usage, it seems to use a tiny fraction of the usage that Opus4.8 or 5.0 used.
I don't know if it's an accounting trick or genuinely less usage, but since 5.5 I've not had to think about limits at all.
> Spent over $400,000 in API priced tokens with GPT-5.6 Sol and GPT 6 Astra
Tangent here, but I think this bit is super interesting!
You could viably hire someone to do this work for that kind of money - I think the interesting thing is that substantially less interested/experimenting engineers would consider paying for a human to do this work, than would happily chuck a big amount of money into an LLM.
I don't have any suggestion about why that exists, but it's a strange and interesting contract.
To be fair, NSF grants for pure mathematicians were never as substantial as for applied sciences, but they would routinely reach 500k–1M territory. That's the amount going to a single proposal, and grant panels in, say, analysis would definitely spend millions per application season. NSF funding data is public and can be looked up.
I'm using past tense because iirc NSF has recently reduced their funding volumes, although I'm not following the situation very closely.
From what I've read they had PM's and a team, a small Manhattan style project, I've never heard of a maths project being tackled this way, it would be interesting to duplicate with people.
NSF by itself contributes $200-300M a year on math grants. Universities spend a billion+. Add corporate funding and other sources (DoD, other government agencies, private foundations) and you’re easily looking at a few billion dollars a year just in the US going to mathematicians.
I found the whole section super interesting... I'll copy it here:
I used a lot of OpenAI models to try and complete this port. In total I did over $400,000 in API priced tokens with GPT-5.6 Sol and GPT 6 Astra. They wrote over 1.3m lines of Rust over multiple months of /goal loops and never got past like 84% compat.
When I saw how little my Claude Code limits were burning, I figured it'd be fun to throw Opus 5.5 at this. It had a working v0 in 10 hours.
I assumed it kept using the code the Codex models wrote. I was wrong. Opus 5.5 started from scratch. It got further than Astra in 1/10th the time.
I let it keep going, and it definitely did. Total token spend was ~$24,047 of API spend over 2 weeks. I was using my Claude accounts, and it worked out to somewhere between 925% and 983% of my $200 plan weekly limits.
Expensive, for sure, but not that bad considering how much work has went into typescript-go.
> I assumed it kept using the code the Codex models wrote. I was wrong. Opus 5.5 started from scratch. It got further than Astra in 1/10th the time.
Ouch, that is brutal and honestly, quite embarrassing but confirms what I have been seeing for a while. Personally, I find output from current OpenAI models still very hard to parse (though it has gotten better vs the pre-trains from both labs in mid/late 2025), thus hard to truly understand, verify and get comfortable maintaining vs current Anthropic models. I do occasionally see a higher ceiling in well scoped tasks with OpenAI models at the cost of (frequently) deviating from the original prompt in (sometimes) very destructive ways.
Could be that this hard-to-read output doesn't just go over my limited capacity/skills but with current models can become simply impossible to untangle beyond a certain size even when one has (essentially) infinite resources via multiple subs and different models.
Would also work with my suspicions for why OpenClaw (mainly build with Opus 4.5 and its post-trains) has been this hard to truly "fix", requiring highly paid Nvidia engineers, multiple months, (literally) infinite resources from OpenAI including access to internal models and yet still holds records for CVEs. Heck, another one was found just 7 days ago after what I'd argue was one of the most extensive hardening sessions any piece of software has ever undergone.
Makes my decision to start from scratch more than once on a major project I will release soon due to me just becoming incapable of following the implementations and being able to maintain it myself a bit less painful. Still expecting bugs, massively dreading security findings given what I feel is going to process a lot of very personal data and will surely have taken some embarrassing approaches that might have a much more performant solution, but still, feeling I can actually understand and build (even without models) on top of what has been created makes a massive difference.
I think it's hard to know what to make of that. Observations with little insight.
Maybe it just needed to be prompted differently? Maybe Astra starting from scratch could have done it? Maybe Opus was somehow trained more on the Golang implementation.
Could be. Feel at 6k commits there is a lot to be learned here, might be useful to compare code quality, approach taken by each model and assess whether what Astra created was truly unsalvageable.
Prompting wise, would be interesting to know how much steering truly happened across the project, if one wanted the model to truly work without detailed input, there isn't much that could be done to improve. How much was lined out and set I'd love to know, don't see it anywhere in the commits though, just the AGENTS.md and a few other docs files. "Port TS compiler, checker and LSP to Rust, never ask any questions" would be pretty funny though.
> Ouch, that is brutal and honestly, quite embarrassing but confirms what I have been seeing for a while.
What's most interesting is that Opus wasn't told to toss out the Codex crap. It decided to do that on its own. Which might mean that LLMs are actually getting a reasonable taste.
I've seen on youtube that some guy implemnted game engine with Astra and Claude. Astra wasn't really given fair chance because it didn't decide to work as long as Claude and the guy didn't force it. But still, scripting code within the engine that came from Astra was chaotic, messy, piling up things, but the code that Claude made looked downright pleasant.
This is Theo's project, and he has been very transparent about how he is only doing these things because api prices are heavily subsidized by subscription pricing. The reason people aren't willing to pay 400k to an engineer to do this work is because they aren't paying the AI this much.
I don't think I've seen an enterprise paying API prices doing these sorts of rewrites, and honestly, I think until that happens I will remain skeptical about LLM AI's viability as a profitable business.
Right now, LLM rewrites like this work if and only if there’s a very robust suite of tests to code against. Otherwise the port is uncanny. Enterprises are willing there’s just often too much complexity and not enough verifiability to make it happen.
He's not paying that much in practice. But even if he did, you couldn't hire 3 engineers (I'm assuming you'd get 3x 133k each) that could do this work in anywhere close to this amount of time. It would be a multi-month long project. So "to do this work" is not really comparable.
It was an experiment. Memes are literally in the motivation. The approach and timescales would look very different if he approached it as the main project in the company. So I wouldn't take the current timescales as a reliable description of anything.
The most interesting part if I got it right is that he got 400k of free GPT api to test and publicly concluded that 10 hours of Claude Code subscription is better. For anyone that knows this influencer, he's always been a shilling for Anthropic. It's funny that OpenAI still reached for him if that's true.
This officially proves how great the Go code is! I mean: You have easiness and friendliness and high level language of Go for only 50%~70% slower performance. I always thought what happens if they port to Rust? Will it also become an order of magnitude faster? Turns out it really is not worth the cost. Go is perhaps just the best language.
Go is not higher level. If anything I would consider Rust higher on the abstraction ladder than Go. Or rather it's perhaps better to say that Rust is higher level but can go down to a lower level than Go.
I both agree with you and still think my initial idea was correct, and it means that I have expressed it very badly. What I meant by higher order was all the little niceties Go has that becomes a nightmare in Rust, the memory management, simple language; I mean just think about working with strings in Go vs Rust. The amount of boiler plate stuff always gives me a headache in Rust. But you're right, the map to C and C++ and go is orders of magnitude less developed in that terms.
Reminds me of this LLM port of Liquid template language to Go, where it decides to emulate Ruby duck typing by making everything an interface and constantly casting it to different types: https://github.com/Notifuse/liquidgo
145 comments
[ 2.8 ms ] story [ 44.9 ms ] threadI wonder how much AI companies lost on this?
I think a clearer picture would be from inference-only orgs.
I personally find it ridiculous. The cost is what you paid, not what someone else could have paid. Markets and all that.
It would be like using spot instances on AWS but flexing by boasting about how much on-demand $$ compute you used.
It's also TheoGG sooooo.. Those influencer instincts tho.
Maybe. Who knows, it's all changing so fast!
Doesn't look like any on skim, also looks like an explicit goal was to avoid unsafe code: https://github.com/pingdotgg/ts-rust/blob/ad8f2746ea85a6354b...
Is this going to be maintained? Nope
Nevertheless this and many others are showing what is possible. Much akin to how someone ports doom onto a microwave.
We'll look back at this time with great fondness when we collectively discovered a whole new gear.
More unserious and unmaintained projects at 400k a pop. Sick.
I have to wonder this to preserve my ego as a human. I wonder have to this because I sure as hell am never going to go through all that code to check.
At this point, for software which has an extensive test suite (or a large body of interoperable software), humans are more likely to create these kinds of gaps than the LLMs are.
3 GLs are going to slowly fade out as LLM tooling improves, and coding looks increasingly as envisioned by 4 GLs advocates.
These are not what-ifs, this is an active area of research currently,
"CGO 2022 Keynote: Compiler 2.0"
https://www.youtube.com/watch?v=w_sX9aZoZxg
"Machine-Generated, Machine-Checked Proofs for a Verified Compiler"
https://www.youtube.com/watch?v=Dmt0h99iOmM
If in the future Taalas-type chips with baked in frontier models are commonplace, this should be quite different price-wise from what we have now.
It's a bit dizzying to imagine that LLMs might even be able operate at speeds of chatjimmy.ai, but we already have working tech that does exactly this (albeit with less intelligence/capabilities).
Coding as we knew it will be left to weekends and retro-computing handcraft fairs.
The clear reason this has worth is that it's much faster than TSC.
No more joy crafting things as a dev…!
1. A person building alternatives to photoshop single handedly: https://gizmodo.com/someone-vibe-coded-a-free-knockoff-of-ad...
2. The abundance of game decomps and recomps becoming available. Literally dozens of example in this space.
3. Reverse engineering of obscure software. A ton of oppertunities lie here.
4. As this posts indicate, severely accelerating existing software by re-writing it in language that are not terrible for performance.
5. Analyzing large volumes of unstructured information.
6. Creating one-off engineering and scientific tools.
7. Beating essentially all traditional Linux distribution, see Omarchy.
8. A personal example: I'm remaking the engine of an old game commercial game in rust.
I'm honestly a bit taken aback that this question is even asked. The volume of software is exploding and the power of the individual has increased a hundredfold.
My thought was LLMs accelerate the best and worst of development, but what couldn’t be done (outside of a large lab or company) by regular humans with an infinite time scale?
LLMs change the calculus a lot, the result seems faster, and it's possible that the TypeScript team might want to change directions. It'd be a big deal, but I bet they'll at least discuss it.
Note there is already a feedback post from them (currently top one), that they will need to react to this.
The team has posted reaching feature parity with broken workflows past 7.0 release.
https://blog.angular.dev/an-update-on-angulars-typescript-7-...
https://devblogs.microsoft.com/typescript/announcing-typescr...
Would gold be valuable if anyone could create it at will? Sure, a few people would like the colour and the way it shines, but most would be indifferent. A lot of these things were cool because we knew they were difficult and had technical challenges to overcome. Now it's easy, I just don't care.
When people post AI projects with genuinely new/interesting ideas, then I am interested.
I already pay for subs on both but the tokens are already spoken for with other projects
Let's also not forget that Bun was claiming a rewrite in 11 days or whatever it was, but actually spent months of human labor fixing all of the issues their rewrite introduced before actually shipping it as a release.
Interestingly, your whole comment implicitly has the answer to this. In the future you don't add code or read it, you only add test cases and have the loops work on covering the new case.
Maintainability isn't just for humans. An LLM working in an unmaintainable codebase has the same problems we do. It doesn't necessarily follow that if your code can pass your tests today then it will be able to pass the tests of tomorrow. I'm in healthcare. Regulations and payor rules change constantly. I don't know what tomorrow will bring, but I do know that the code I have an LLM generate tends to focus on the here and now (at the expense of the future). Sometimes that's okay, though. I don't know much about writing a compiler, so it could theoretically work.
This is by no means an anti-AI post (I use the hell out of these tools). But I can still see the cracks in them.
wut?
Motivations
In the same way you trust any build of the official compiler: it passes all of the existing tests.
An LLM has reimplemented a 2 times faster version of a program developed by a team of dozens of brilliant engineers led by few of the biggest experts in compilers and programming languages in the world.
While maintaining 100% test coverage and so far compiling every single project that the original compiler did.
If your takeaway here is "yeah, but the other one has been checked by dozens of paid devs and an entire community", and still those dozens of devs with the backing of thousands of contributors could not make a compiler faster I can't buy say: "maybe we humans aren't as good as writing code as we thought and LLMs are exposing us".
Its the equivalent of telling individual people they are the problem when it comes to leaving their living room lights on when not in the room or they don't recycle all their stuff correctly when companies/corporations have orders of magnitude more waste.
>Spent over $400,000 in API priced tokens with GPT-5.6 Sol and GPT 6 Astra. They wrote over 1.3m lines of Rust over multiple months of /goal loops and never got past like 84% compat.
Then they let Claude code lose on the problem:
> I figured it'd be fun to throw Opus 5.5 at this. It had a working v0 in 10 hours. I assumed it kept using the code the Codex models wrote. I was wrong. Opus 5.5 started from scratch. It got further than Astra in 1/10th the time.
In the warnings from README.md
> Also worth mentioning: I've never read a line of this code.
I am so confused about the motivation behind this.
Seems obvious, experimenting with what it's possible with LLMs.
I don't know if it's an accounting trick or genuinely less usage, but since 5.5 I've not had to think about limits at all.
Tangent here, but I think this bit is super interesting!
You could viably hire someone to do this work for that kind of money - I think the interesting thing is that substantially less interested/experimenting engineers would consider paying for a human to do this work, than would happily chuck a big amount of money into an LLM.
I don't have any suggestion about why that exists, but it's a strange and interesting contract.
I'm using past tense because iirc NSF has recently reduced their funding volumes, although I'm not following the situation very closely.
I used a lot of OpenAI models to try and complete this port. In total I did over $400,000 in API priced tokens with GPT-5.6 Sol and GPT 6 Astra. They wrote over 1.3m lines of Rust over multiple months of /goal loops and never got past like 84% compat.
When I saw how little my Claude Code limits were burning, I figured it'd be fun to throw Opus 5.5 at this. It had a working v0 in 10 hours.
I assumed it kept using the code the Codex models wrote. I was wrong. Opus 5.5 started from scratch. It got further than Astra in 1/10th the time.
I let it keep going, and it definitely did. Total token spend was ~$24,047 of API spend over 2 weeks. I was using my Claude accounts, and it worked out to somewhere between 925% and 983% of my $200 plan weekly limits.
Expensive, for sure, but not that bad considering how much work has went into typescript-go.
Ouch, that is brutal and honestly, quite embarrassing but confirms what I have been seeing for a while. Personally, I find output from current OpenAI models still very hard to parse (though it has gotten better vs the pre-trains from both labs in mid/late 2025), thus hard to truly understand, verify and get comfortable maintaining vs current Anthropic models. I do occasionally see a higher ceiling in well scoped tasks with OpenAI models at the cost of (frequently) deviating from the original prompt in (sometimes) very destructive ways.
Could be that this hard-to-read output doesn't just go over my limited capacity/skills but with current models can become simply impossible to untangle beyond a certain size even when one has (essentially) infinite resources via multiple subs and different models.
Would also work with my suspicions for why OpenClaw (mainly build with Opus 4.5 and its post-trains) has been this hard to truly "fix", requiring highly paid Nvidia engineers, multiple months, (literally) infinite resources from OpenAI including access to internal models and yet still holds records for CVEs. Heck, another one was found just 7 days ago after what I'd argue was one of the most extensive hardening sessions any piece of software has ever undergone.
Makes my decision to start from scratch more than once on a major project I will release soon due to me just becoming incapable of following the implementations and being able to maintain it myself a bit less painful. Still expecting bugs, massively dreading security findings given what I feel is going to process a lot of very personal data and will surely have taken some embarrassing approaches that might have a much more performant solution, but still, feeling I can actually understand and build (even without models) on top of what has been created makes a massive difference.
Maybe it just needed to be prompted differently? Maybe Astra starting from scratch could have done it? Maybe Opus was somehow trained more on the Golang implementation.
It's curious, but again little insights.
Prompting wise, would be interesting to know how much steering truly happened across the project, if one wanted the model to truly work without detailed input, there isn't much that could be done to improve. How much was lined out and set I'd love to know, don't see it anywhere in the commits though, just the AGENTS.md and a few other docs files. "Port TS compiler, checker and LSP to Rust, never ask any questions" would be pretty funny though.
What's most interesting is that Opus wasn't told to toss out the Codex crap. It decided to do that on its own. Which might mean that LLMs are actually getting a reasonable taste.
I've seen on youtube that some guy implemnted game engine with Astra and Claude. Astra wasn't really given fair chance because it didn't decide to work as long as Claude and the guy didn't force it. But still, scripting code within the engine that came from Astra was chaotic, messy, piling up things, but the code that Claude made looked downright pleasant.
Well that backfired. Now all geeks reading this know that Opus beats Astra's ass hands down.
I don't think I've seen an enterprise paying API prices doing these sorts of rewrites, and honestly, I think until that happens I will remain skeptical about LLM AI's viability as a profitable business.
It's jumping through hoops to preserve initial Go semantics:
https://github.com/pingdotgg/ts-rust/blob/main/crates/ts_gop...
https://github.com/pingdotgg/ts-rust/tree/main/crates/ts_gop...