I bought $18 GLM official subscription yesterday (5.2, but new model version was already leaking on some docs), set it up with Claude Code harness... and I’ve bumped to $80 plan almost immediately. It’s the first model that agreed on a proper security research (red team scenario), executed it seamlessly, including 0-days in WP plugins, RCE, 6.8 kernel exploit adaptation, etc - while playing against another GLM agent as a defender (following HF story)!
I understand that such models can be used by malicious actors, but it’s fair to have it publicly available (and play on your side in case of emergency). This is what changes the world in a better way, I think, not the guardrails.
At my work I have a $500 monthly AI budget. I have been using the $200 Claude subscription and most of my use is with Claude code. I think I'm going to switch to either kimi or glm and use the opencode harness. Both fable 5 and opus 5 have outright refused things like security related bug fixes and making monitoring tools. I am so happy that open models are good now
My beef with the Fable refusals is that it seems to just be flagging keywords, and also seems it flags on keywords the model itself introduced to the context.
In a normal Chat with Fable, something like "How can I exfiltrate a guy from a sticky situation?" reliably downgrades, leading me to believe that Fable just outright refuses once it sees the word "exfiltrate". When it writes a service and names it CloudExfiltrator, the next turn downgrades to Opus.
Opus doesn't appear to refuse on simple keywords, but it does seem like Fable's reasoning introduces enough nefarious-sounding context that Opus will then refuse, and I'm stuck playing the new session game despite having done everything correctly myself and having a totally innocuous prompt. At one point, Opus was happy to continue while outputting commands for me to execute on its behalf, but flatly refused to execute them itself through multiple new sessions. To its credit, it openly acknowledged how ridiculous that was and was apologetic for the safeguard.
I'm open to the idea of some kind of guardrails, but if Fable is so dangerously intelligent as to require the guardrails you'd think they could come up with something a little more nuanced than a list of bad words. As far as I can tell, they've also not done anything towards improving the situation since the model was released, despite the "deliver more capabilities faster" claim.
I generally use the $100-200 Codex/Claude subs, and have been blown away by the usage I get from OpenCode Go at $10/mo. At a minimum, excellent for automatically piping reviews to from Codex/Claude.
I run my OpenClaw on whatever is the latest GLM model and ever since the release of GLM 5 it has been a smooth ride. The models solve whatever problem I throw at them and the code is good enough that I barely ever have to look at it (to guide the mode). The 5.3 release seems particularly strong, I asked if to audit all the scripts that the previous versions have written and it identified some issues and hard to find bugs.
At work, as an experiment, I used GPT 5.6 Luna + Deepseek 4 Flash for a week (I have an unlimited, "within reason", budget at work so normally I just use Fable and Sol) and it's been perfectly fine.
These models take a bit longer (more turns) to solve problems so they feel a bit slower but the end result is often just as good or nearly as good. Because they're so cheap you can easily run multiple sessions in parallel so it doesn't really matter that they're slower.
I've done a few experiments where I've split my terminal in 4, launched 4 clients (each with a different model, including Fable and GPT 5.6 Sol) and compared the output. For simple and medium complexity work open-weight models are incredible effective.
I can highly recommend the 10 USD/month OpenCode Go subscription. It offers pretty amazing value for the money and is a great way to experiment.
what is a harness? The comments below are mixing IDE/ADE but other suggestions are purely terminal things and I don't get what their value is over just a terminal. Is a harness like a loop where it's just a vague thing that everyone nods about but everyone is nodding at something different?
My understanding is that the harness is the set of function calls (or tool calls) that let the LLM interact with your codebase. It's independent of the IDE or CLI.
The tool calls will be, among other things, something like ReadFile, RipGrep, PatchFile, Shell.
When people talk about the value of different harnesses, they're also implicitly talking about the quality of the system prompt.
The same exact model, when given a different set of tools and a different system prompt, can behave differently.
The harness is the agent. LLM's can be asked to output things in JSON for example. The LLM then literally asks for things like "execute this cmd" or search/replace this string. The LLM outputs text, but in a deterministic format that can be parsed. The harness calls the LLM, exposes tools, executes tools the LLM asks for, gates tool use based on security controls. It's the runtime that the agent uses to do work.
The LLM is the core model, but the harness has the prompts/tool definitions, guidance/recovery/correction code. The harness itself is the agent, because same model may perform vastly differently on different harnesses. Agent is the system working as a whole, harness+llm.
LLM is just the thing generating text. It can generate a book, an article, a paper or a conversation. It's the raw "autocomplete" mechanism. You give it input, it generates output.
The agent/harness is the sotware that leverage this "dumb" autocompletion engine to do useful things by sending the good input to the model and doing useful things with the output.
LLM is the digital brain. Agent is a software robot with "sensors" (tooling to read data from files/web) and "actuators" (tooling to perform changes in the environment) that is based on the digital brain.
think of it like the words "car" and "engine", one goes into the other and both have their impact on overall capacity, feel, performance etc. each in their own way. The llm is an engine, the harness is the rest of the car.
A harness is the tooling around an LLM that uses its responses to do work. For example, let's say a harness sends your message to the LLM and the LLM emits a tool call - the harness is responsible for processing that tool call, executing the tool, and then sending its response back to the LLM to get more of its turn. The harness is responsible for keeping the conversation history to send back to the LLM each time it needs more completion. The harness is responsible for displaying the user interface that allows you to send instructions to the LLM or configure MCP servers or whatever. Claude Code is a harness, Pi is a harness, OpenCode is (technically) a harness, it's basically the LLM "client". The model is what generates the completions itself, but the harness is how you invoke it. And generally when people say "agent" they mean the harness hooked up to some model.
You might be joking, but a harness provides much more than just the prompts: at a minimum, it provides the system prompt and the built-in tools that the LLM can use, but it can also provide things like subagent management, custom compaction logic, session forking, etc.
It’s a new buzzword where everyone says it but means entirely different things. You can use it to sound smart whenever there’s a conversation about how you’re not extracting value from your coding agent and also when you are extracting really good value. Very flexible term!
I also quite like driving OpenCode through something like Kepler / Paseo and tools like that (with those I can still use my Anthropic Condition by Claude Code being treated similarly - as something that gets tasks dispatched to it, while the GUI I see is Kepler / Paseo).
On the desktop side, ZCode was surprisingly usable for something that came out of nowhere (I wasn't aware of it at all before trying out the GLM Coding Plan): https://zcode.z.ai/en
Last time I tried some of these, none of them had the "manual mode" that CC has, where it shows you change by change as diffs and you can edit them before accepting and moving on to the next change. I like that because if it's going off pattern I can spot it early on and guide it correctly, instead of having to review the whole completed diff at the end when it's too late. I should spend the weekend checking them out again to see if they added that but I assume with everyone going full agent mode they probably didn't.
Both Pi and OpenCode let you customize them. You tell the AI you want "something like claude code manual mode", and they'll modify your configs to do the same thing, or build an extension for you
(however, it's much faster to use Plan Mode to build a plan of what it will do, and then execute the plan in Build Mode. you can also have the AI make a script that will be executed deterministically)
The philosophy with Pi is it is minimal (but functional) out of the box and easily extensible. I'm not familiar with that feature but I would not at all be surprised if someone already coded a Pi extension that does it.
The out of box experience of omp.sh is wow imo so much nicer than Claude. Claude spends too much time being nice and gassing me
Up and omp just gets to work. It’s idk smarter like a far better system prompt and all around loop.
It's less about malicious intent and more about predictability.
When I install software on my computer with apt, I trust that all the files will go to the right place and install scripts are going to do sane things relative to the rest of the system. And I can just uninstall the whole thing with one command later if I so choose.
If I curlpipe a script, I get none of those guarantees. I have seen curlpipes that put files in weird places, guess the wrong OS, and mess with config files that I didn't want them to touch. When they break or I want to uninstall, I have to sit down and understand a (possibly minified) script to clean things up manually.
Yes containers are a half solution to this, no I don't want to use containers 100% of the time.
its a good way to check if people are insane though. would be a cool tactic for new hire evaluation, monitor them setting up dev environment. do the curl | bash, and its instafail
In countless corporate environments (including in highly regulated industries), far from being a firable offense, piping curl to bash is often a prescribed step in setting up the standard dev env. The cognitive dissonance is soul crushing. Maybe they're testing for one's ability to tolerate it.
T3 Code has been amazing. Completely free. Really impressed with the desktop app and the mobile app experience and the way it works seamlessly has me actually accomplishing tons of stuff while I'm out on mobile that I would otherwise have to wait to come home for. First time in a while I'm actually excited to use a desktop UI instead of the terminal. Blows away the official Claude Code mobile app. I can switch between my Claude and Codex monthly subscriptions in it as well. There's a TestFlight beta SwiftUI mobile version that's so much nicer than the one in the App Store. I'm running the nightly version of the desktop app.
And this is coming from someone that's not particularly a big fan of Theo. T3 Code should get more recognition; people aren't just aware of it yet.
It indexes the code efficiently, seems to find stuff quicker, it has a very nice UI (much better than Claude Codes IMO), it has a nice sub-agent UX which I find triggers more reliably, diffs render nicely. Otherwise it just seems to work in a purely vibes sense.
That said Claude Code is perfectly fine. I just prefer the integrated experience of using Cursors since I already use VSCode, but I still mostly use Claude Code because of their Max/Fable plan.
It’s not just bloat at this point. I run oMLX and run models locally. using Claude code on the first message dumps 40k of tokens that my laptop takes 5 mins to compute.
Piggybacking on this thread to ask my question: What are alternatives that are multiplayer (team oriented) by default? For example, I want my team to see all my sessions easily, vise versa. another way of stating: all the agents are running in a container that that any member of the team can view and interact with.
Just as a +1 anecdote. I enjoy using pi a lot. I used to h think the harness matters a lot but with the current iteration of models I am starting to sway that while it matters it’s less and less important and that CC is bloated. I did some quick tests when I switched and a task that would take $5 in tokens would be completed in $0.50 in pi. Very anecdotal and I don’t have a test framework setup to make this very official but increasingly felt like CC was spinning its wheels on the easiest of tasks.
The token cost difference is pretty interesting. I wonder how much of that is the harness itself versus how aggressively each one loops, plans, and calls tools. A proper apples-to-apples test would be really useful here.
I’ve tried a bunch of them, and I seriously do not understand these recommendations. It was a rough road and a steep hill, but right now CC is absolutely the best harness on the market, as for me, whatever top tier model is under the hood (mostly, some of them, like DeepSeek, don’t fit CC at all).
Inversely I don’t understand the praise for CC. These days it feels like bloatware. It absolutely can get the work done but when I measure on token and time use it ends up being a multiple of pi like harnesses.
CC works but for me it felt like increasingly they have zero incentive to make it a great experience. You hear folks like Boris talk about spinning up thousands of agents over night and agents chatting back and forth in GitHub issues and while I think it’s great from figuring out what the future looks like I don’t think it represents the reality of ROI today. So the folks building the tool are so disconnected I am simply not sure it’s a great experience anymore.
So is the quantitative difference in token use the only difference or do you think there's also a different qualitat? I'm on CC only and immensely happy. Very productive both at work and privately and at work I average around $250 a month which probably means nothing but it's little compared to my salary.
For me, at least it's that the newer Claude models seem optimised for one-shotting things, which is not what I want. As the amount of code per turn increases, I have a harder job keeping up and ensuring that it's doing what I want.
That being said, I had to nope out of a similar thing from GPT 5.6 today, so it appears to be a US frontier lab issue. Claude is particularly bad though, as it produces far too much code even when I tell it not to, unlike GPT (and Kimi) which at least listen to me a little better.
More generally, I want a usable human review experience, and Claude code doesn't deliver that for me.
Quality is hard to measure and I would not say the concern is so much cost but the intersection of cost and time. Often I am jamming on something and I like being somewhat in the loop. So maybe same level of quality, I am using Anthropic modela for both harnesses, but I get to the output quicker and at a drastically lower cost.
It's buggier for me than it has ever been before. I don't think that agentic coding always leads to such a buggy mess. I just don't think that the Anthropic front-end software team is very good at agentic coding.
It might be time for the Anthropic devs making 500k a year plus more than that in options to actually write code again? I can’t believe that humans in meat space are reviewing all that code. Maybe that’s why Claude code is feeling a bit bloated and the models are a bit lacking of late is too much of the bot is writing the bot.
Funnily enough, I would say almost the opposite. CC’s feature set is basically table stakes for an agent these days (does it have ACP yet? Very close to behind table stakes if not) and it has a lot of bloat powering that.
IMO part of it is that the underlying LLMs have gotten better enough that harnesses feel better even if they haven’t changed. I have a toy harness that barely implements the features you’d expect and it works surprisingly well. Like there’s literally nothing clever, it calls tools and that’s about it, and it still mostly does the right thing.
Why would anyone ever need ACP? I'm not trying to be an asshole. I just seriously don't understand the value proposition.
Edit: lol, I don't think ACP is even actively developed anymore. It seems to have been merged into another seemingly pointless standard with an even worse name, A2A. [0]
Based on the fact that Claude Code is only optimized for Anthropic models, whereas Pi and Omp are optimized for a wide variety of models, including open weights.
Can't edit my post anymore, but here's the omp blog from February talking about improved tool calling rates across 15 models, with only the harness being tweaked to get the improvements.
What’s the counter argument? pi and ohmypi are pretty fantastic. Of course like all developer tools it depends how you do your work but I am not sure what you are trying to achieve in your comment.
how would i comeup with counter argument if i dont know what original argument is. No one is disagreeing with your subjective experience, gp comment said 'better' without qualification.
They're invaluable for developers to fix their code. This is definitely an area where AI decisively beats human devs in a very valuable way. It can try so much surface area so fast.
If it won't attack my stuff, it won't help me build my stuff to be secure.
Yes, I am tired of Claude and GPTs. I am ready to diversify my $300 per month on other vendors. Will try GLM. How was your rate limits and availability experience on $80 dollar plan?
It’s comparable to Anthropic usage, to be honest. 2x GLM agents ate 18% of weekly usage on this mid-tier plan within ~8 hrs (non-stop work, a lot of tool calls, appx 4 compactions each), I think. I didn’t make a proper statistics snapshot, sorry.
> I understand that such models can be used by malicious actors, but it’s fair to have it publicly available
I feel like there should be some mechanism to prove you own the code/app/site/whatever and it will remove the guardrails from the LLMs allowing them to find and fix these vulnerabilities.
Don't we already do this with services like Let's Encrypt, which is arguably more sensitive? If you had the codebase you could fake it, but it would still provide some amount of protection against abuse.
With Let's Encrypt, all the verification is done on their side with them controlling the connection between themselves and whatever they're trying to verify.
In this case, you can put whatever you want between the harness you're running (or modify the harness itself), and essentially "lie" to the model. Any verification technique would be fairly trivial to bypass, while you continue to run the harness locally.
Yeah, for web apps, you can trick models by simply proxying it and pointing the models to that localhost. They then think they're not working on a live target.
Have personally tested this with Opus and Sol and it works.
Classifiers are tricky though. Here's where open weights will win.
How much usage do you get out of it per week? How many millions of tokens?
Anthropic was stingy as hell with its Fable and cybersecurity nonsense, switched to OpenAI which is much better but still not enough. I'm tempted to switch again...
No no no, absolutely not. There's only one man that should be allowed these privileges, and his name is Dario. Dario alone can deliver us to salvation. The lord himself shalt smite these companies and models from this barren earth, and Dario will rise from the ashes to ascend to godhood. Dario. He alone has the power to decide what capabilities us mere mortals have access to.
Nope, they want shared artifact(s) between web and desktop (not sure what do they have in mobile apps) clients (Chat/Cowork/Code/CC-cli), so TS is the way to go. It’s a reasonable choice made by engineers.
Is it some Claude Team/Enterprise only problematic? I'm using two 20x Max accounts almost non-stop (Fable/Opus) for 1.5 years at this point, zero issues with both client and infra sides (from US and in travels). When I'm reading such messages it feels like either I'm lucky or it's a part of some campaign.
I’m on the biggest max plan. It is riddled with annoying bugs for me, only been using it for a little over a month. Settings screen flashes randomly. But most annoyingly: sometimes when forking chats or sometimes for no reason, the UI just straight up eats my previous messages. The model is still aware of them and can recount them if I ask but the visible history is gone. And that’s not even all of them. Fable 5 is just too good that I put up with it but it seriously raises concerns for me that even with infinite compute these companies can’t even deliver a functional chat UI.
I've been on one to two of these plans for eight months or so. There used to be a lot of issues with CC's terminal but at least in iterm2 they have largely been sorted.
I commend you for burning 6 figure losses into Anthropic thanks to their subsidies singlehandedly but I am curious to learn what you've built with this so far, not to be snarky, I just wonder what people actually produce while having these run nonstop.
No idea but we have few Team Premium seats and everyone is encountering issues daily for past two weeks. From straight up outages to vscode extension/CLI refusing to process messages. It's been unusable for most of our work hours past two days.
Over the last 1.5 years, they had a few failures with their auth system; two or three times auth failed for a few hours so I could not work. But otherwise they have been just fine. Some issues, but not significant.
I think you’re just lucky. Look at the Claude status page to see just how often they have outages (it’s almost daily). Even most of the green days have issues if you hover over them, they just don’t count them as outages.
If I understand correctly, time spent within this interstellar ship reference frame would be significantly less than 75k years. And at 1c speed, 48 years on Earth would be ~2 years of "flight" for the passengers, correct me if I'm wrong.
You are not wrong, but it still wouldn't make a whole lot of difference. First of all with any technology we currently have, including hypothetical proposals, we can't even get close to 1c. The best we can do, and this assumes purely hypothetical unproven ideas, is accelerate to a fraction of the speed of light. Acceleration itself will take up a significant amount of time and even then we are nowhere near the speed of light. Project Orion would in theory be able to do 3% to 10% of the speed of light. That's about the only proposed propulsion we actually have the technology for in the sense that we might be able to build it this decade.
So let's be optimistic and assume we can reach 10% of c (unlikely), that is 10 years for one light year, meaning that it takes about five centuries. Practically speaking 10% of c is unlikely, it takes about a month to accelerate to those speeds you'd need an insane amount of mass in the form of bombs to explode to just get up to speed. Which is why 3% is more "realistic". Oh, did I mention that there is no realistic way to slow down either? In theory it is just detonating the same amount of bombs in front of the craft, but in order to take that mass on your journey you'd need a lot more mass to get up to speed and the numbers get silly rather quickly.
So "realistically" at 3% of c we are talking about roughly 1500 years to just speed past our target.
Also, the original remark was about sending probes, not humans. But at 3% or 10% of c time dilation isn't much of a factor either so you'd need a generational craft if you want to send people.
Man, that’s not how intelligence works. Eventually if a civilization doesn’t blow itself up, it will achieve anything, and it’s not incremental steps. Speed has never been incremental steps. in 1810 you would’ve been thinking about faster horses or some shit like that. Def not rockets. OP said hundreds of years. In hundreds of years from now our rulers will definitely have that kind of tech, whatever tech it is.
Often technology is approaching an asymptotic limit that it will not go beyond. Know about a hand waving will change the laws of physics. There's every reason to believe that the speed of light is a maximum that we will never be able to choose because it requires infinite energy. While it is true we don't know everything, and there are certainly parts of it we don't know, the things we do know will have to obey and predict the things that we currently do know to be true. The laws of relativity, although annoying, are very good at describing things. If anything replaces relativity, it's going to replace it while still predicting exactly the same things that relatively does that we have verified. That doesn't leave much room for a theory that might and faster travel.
Is there a way to turn options-based AskUserQuestion off? I couldn't find it at all, and the "options" selector is the most annoying thing in CC for me, plain-text is the only way, everything else is distracting. I know, I can use `n` (sometimes) or cancel it, but both bring more pain that just regular communication (cancellation does one more chat step + requires an extra action, same with `n`)
@anthropic, can you finally add $800-$1000 per month plan and allow us to work instead of tracking your weekly changes and dramas? I think, we (individuals, small-medium biz, first of all) did our best to help you train the model like Fable. Enterprise-level lockdown (and API costs define this) is... unfair? I mean, we all knew that you all will just use us, but it's AI, right? For people, right? Right?
The only reason this is happening -> someone (US gov?) decided that it's time to bail out those who would inevitably die within a year or two otherwise, middlemen.
There is absolutely 0 chance Anthropic or OpenAI will get a government bailout (it might still happen in this admin but it makes no sense). These companies are not like banks which are fundamentally important to the economy. Sure AI is important but Google is not going to die. Why would you save OpenAI and Anthropic when google, amazon or microsoft can just gobble them up when/if needed?
The average taxpayer gets 0 benefits from LLM. It might change in the future but for now that is true. This was exactly the reverse with banking, everyone would lose their own money if the banks just disappear tomorrow
If you look at it from the perspective of the current US administration, they see that almost all GDP growth in the past year(?) has been related to data center growth. If all of a sudden that industry is gone, you're looking at GDP stagnation or drop that looks terrible for the current party, hence the potential for a bailout.
Personally I hold the opinion that the investment into data centers would shift into something else, so no real GDP drop, but I'm not sure that's a certainty the same as 'bailout keeps the current story going'
Imo the data centers is where things start to get scary. Anthropic and OpenAI aren’t themselves indispensable, but when all of the industries downstream of them taking on more and more debt to supply projected usage in the future I could see the USG forced to bail someone out, if only to shore up their creditors.
That's not a bailout for AI labs, I meant the "bailout" for Salesforce and others. There's absolutely no place for them in the world where we have Fable+ models. For many of them. Most of them (we just didn't get this feeling yet). Someone just trying to maintain the old world order, that's all. I don't think US economy would fail if those absolutely useless giants would go down.
Sure they do. Someone they know is using AI to ask for help with something, which makes their life easier, which makes things easier on them as well, which is a benefit.
I strictly use only local models, so I agree, but the project was built with Fable so my argument hinges on the assumption that the maintainer is going to continue using Fable and needs to pay for it.
Humans have time-cost too, much higher than machines. Considering SOTA right now, for a project like this it would make more sense for the community to contribute and verify tests, sponsor updates with $.
Not convinced. I was looking for an answer like "it doesn't actually have parity with CPython." If it does, that's a decent indication that it can be sustained.
I've been doing some reading recently around what the literature expects a professional software engineer to produce in a day.
In ~1976 Mythical Man Month era it was around 5 lines of fully debugged assembly.
Code Complete 2nd edition ~2004 bumped that up to 10-50 delivered lines of code per day.
I found other estimates of around 20-60. I need to pull them altogether into a cited article.
Based on that plus my own experience I think 100 lines per day of production-level, reviewed and debugged code was a reasonably higher target for a professional software engineer up until just a couple of years ago.
Today I'm frequently pushing 2,000 to 4,000 - and that's not vibe coded junk (I can easily hit 10,000+ if I'm not reviewing anything), that's code that I've reviewed and am happy to put my name to.
Obviously counting lines of code is a stupid, easily gamed metric. But I still think there's signal there. If you want to build a sophisticated piece of software you're going to have to write a bunch of code to do it. Writing at 1,000+ lines of code per day vs 100 will get you there faster.
It's possible, but we're at the moment when most of us can ask Fable to implement a custom compiler to a custom target for our favorite language, and even use it as a part of custom solution. Why do I need someone else's implementation? Where's the magic in this project? What's the secret sauce?
>Where's the magic in this project? What's the secret sauce?
Someone else paying for the tokens.
Also someone seeing it through (should that come). Obviously we're not "at the moment when most of us can ask Fable to implement a custom compiler to a custom target for our favorite language, and even use it as a part of custom solution", without thousands to spare and lots of time to shape the solution.
Even if it does cost thousands (does it? I genuinely have no idea how to scope such a thing) that might be a good price if a custom compiler to your custom target is something you really want. People have paid far more for far less.
If you're a hobbyist trying to compile python to your weird little arduino based thing, then that's a lot of money and you would want to use somebody else's solution, no doubt.
But if you're an aerospace company trying to compile for a flight control computer (and I guess you really want to use python for some reason), spending thousands of dollars on tokens to make and maintain a custom compiler could represent serious savings.
The big picture impact of AI that I see/anticipate the most is SAAS dying out because AI coding makes this kind of enablement and support software easier to make in-house, and this feels like an example of that, but maybe I'm seeing what I expect to see.
Just eight years ago basically nobody wanted to pay for compilers and developer tooling, and now you're suggesting people will spend a thousand dollars for a compiler they'll have to maintain themselves just because they're willing to pay for AI generated tokens but not for finished tools?
>But if you're an aerospace company trying to compile for a flight control computer (and I guess you really want to use python for some reason), spending thousands of dollars on tokens to make and maintain a custom compiler could represent serious savings.
If you're an aerospace company you're willing to pay thousands of dollars for a compiler, because you need a DO-178C certified toolchain so that you can DO-178C certify the whole airframe. Suggesting AI here tells me you have no clue about the realities of aerospace, because you've just thrown out the entire value proposition of the commercial toolchains.
>Even if it does cost thousands (does it? I genuinely have no idea how to scope such a thing) that might be a good price if a custom compiler to your custom target is something you really want. People have paid far more for far less.
I wouldn't spend $100K in tokens to get a custom bare metal Python. Or even $10K.
And I'd guess that most devs wouldn't either, unless they spend $10K like it's nothing.
People that have "paid far more for far less" are people who have the money to buy $10K watches, or fancy multi $1000 clothes.
This isn't about the shell of a calculator though, but the functionality. Like if the only operations are addition and subtraction, theoretically you could derive the effects of other operations but it's extremely limiting.
So yeah, half of Python might still be Turing-complete, but it wouldn’t really be Python for any practical purpose.
Just like how a device that can’t multiply or divide is not a 4-function calculator; it’s more like an adding machine. Many of which did multiply by serial addition.
If you write a program in python, say a hello world:
'
def hello_world():
print("hello, world")
'
Is that not python? Yet it uses a subset of python?
That program can be run by either a python runtime, or a python subset runtime.
Now if you were to run a python subset program, like a hello world, you would get:
'
def hello_world():
print("hello, world")
'
Whoah, it's the same thing.
Turns out every program you write with a subset of a language, is valid for the super language.
Subjectively also, if the subset is big enough, it feels like that language, if it uses 'def' for functions, that's python. 'I know it when I see it' kinda deal.
I think the confusion comes from the mathematical folk reading "subset of X is X", and implying that "subset of X=X". But this is natural language, not mathematical language, when I say that "dog is mammal", I'm not saying that "dog = mammal" I'm saying that "dog ∈ mammal", and "subset of python ∈ python"
Mojo folks created a new language, officially called it "superset", and trying to sell to enterprise. And it's not a superset by definition, because it can't run it's "subset" (the original Python) without CPython (which was used as libcpython under the hood, iirc). It's a travesty.
The “status” section of the project’s readme explicitly says that it is not passing the full test suite, and that the AOT compiler passes fewer tests than the JIT one.
It also explicitly says that they’re still working on building out the standard library.
I’m maybe not as pessimistic as leobuskin, but they are absolutely right that this is not the first time someone has tried to build an alternative Python implementation, and that all previous ones have failed because they weren’t able to get close enough to 100% parity to be acceptable to most users. Python is an unusually quirky language. I kind of wonder if “written in Rust” adds an extra headwind here because there’s nothing even remotely memory-safe about Python’s extension mechanism. I don’t know enough to know, but I have read about the death of a few of these projects in the past and a common theme of the post-mortem seems to be, “It went so smoothly at the start that we were caught off guard how much of a brick wall the last 5% was going to be.”
Your reply would have been much better without the first line [0]
> Please don't comment on whether someone read an article. "Did you even read the article? It mentions that" can be shortened to "The article mentions that"
No, it wouldn't, because he didn't actually read the readme which clearly states that they are still working on passing the CPython test suite and that 5x performance is an aspirational goal, not something they accomplished yet.
>What is explicitly not done yet — this is the active roadmap, in order:
>CPython test suite (cpython-full): the standing grind; failures are clustered and burned down per wave.
>Stdlib build-out: _io/os, math/struct/random, collections/itertools/json, datetime, importlib parity — each lands as a native module plus a differential corpus module.
>AoT parity growth toward the full corpus, plus single-binary product polish.
>No-GIL/free-threaded runtime hardening: thread/GC/signal stress is now on the default runtime path, with remaining gaps tracked by the ratcheted suites.
Overall the substantial parts of his comment are completely wrong and the subjective parts are not much better
>With AI it's 100x easier to maintain than by hand.
This is an unsubstantiated opinion. In practice AI has a limit well below 100x.
>It reminds my on pperl. same approach using crane lift. Looks good
>This program turns ordinary perl scripts into long running daemons, making subsequent executions extremely fast. It forks several processes for each script, allowing many proceses to call the script at once.
Which sounds nothing like pon, which is heavily inspired by bun. Meanwhile if it's this: https://perl.petamem.com/ which took quite a while to find, then I'm wondering why that would have precedence over bun?
Once you add the first sentence, it basically turns into a negative value comment that shouldn't have been posted.
I noticed that it wasn't the best comment, I was only concerned with the tone, and I feel like dang has enough going on that we also need to help elevate the conversation. I admit there's some delicious irony in the accuser committing the same crime, but it doesn't improve the discussion to revel in that.
> What is explicitly not done yet — this is the active roadmap, in order:
> CPython test suite (cpython-full): the standing grind; failures are clustered and burned down per wave.
I understand that such models can be used by malicious actors, but it’s fair to have it publicly available (and play on your side in case of emergency). This is what changes the world in a better way, I think, not the guardrails.
reply