rapind

I agree, and I started sending well defined review packages to Luna x-high instead, which is how I have this setup when using Codex (Sol drives, Luna async reviews, and Sol keeps moving). Same process with DS really dropped my DS usage by a lot and I think that might actually be the secret sauce. Especially if you use Luna through a subscription (probably the entry pro level would be enough). I'm not sure what a Luna equivalent open model is though. 5.6 Luna max was catching a lot of issues while my implementer keeps rolling. Last couple of days I had 3-4 Sol Mediums running with Luna x-high reviewers (async reviews) and one Sol medium orchestrator and Astra X-high to plan everything out.

Gotta be honest though, I don't love fiddling with this all the time. I would rather be working on my projects than evaluating my usage. Having DS in my back pocket should i need it is a relief. The providers get fiddly though too.

That's an unfortunate experience. Think of v4.1 flash as actually v5.0 flash. It's night and day compared to the 4.0 flash (and 4.0 flash was unintuitively better than 4.0 pro). I would re-evaluate with v4.1 flash. I'm not saying it better than Sol or anything, but it's in the ballpark.

Fireworks directly. At the time they were the best value of cost, speed, ZDR. They got slower on me though, but I think they are retooling, so maybe things have or will get better again. I think fireworks is primarily for when you want to do your own training on top, which I wasn't doing.

I found omp used quite a bit more tokens than my fairly basic pi setup... but most of those tokens would be cached with DS V4.1, so maybe worth if there are gains elsewhere.

pi. I wasn't even going that hard. I checked the logs for Sep 18 and I did just shy of 3b input with approx 98.5% cache and 5.8m output, which cost around $35. Most of the was a Rust code review exercise with 1 driving agent and a varying number of subagents (up to 6 some times). I do the same with with Sol med/high driving and Luna x-high reviewing and get at least as much done if not more in a day, but I'd use up two x20 weekly allowances for the week. Worth noting that token cost isn't super meaningfull on it's own, because DS is super token heavy (but also great at caching) compared to Sol. (my stats show DS uses 3x the tokens as Sol)

The shape of my work changes obviously, so it'll vary, sometimes more, sometimes less. For example, fixing all of the bugs and defects I found that week was 2-3 times the effort and chewed through my ChatGPT allowance, but I had banked resets...

Also worth noting that codex models have been kind of all over the place recently with their usage... and it looks like costs are changing again.

So I took Deepseek V4.1 Flash for a spin maybe 2 weeks ago now (before Luna 6 and Sol 6 were announced), and I racked up $100+ in about 2-3 days. It was pretty great, but it uses way more tokens (TPS is fast, but it's way more tokens per turn) than Sol 5.6 which I found to be about it's equivalent at the time (on medium or high, with DS on max). My cache rate was around 98-99%.

It would definitely cost me more per month than a x20 ChatGPT or Claude plan, probably around $400+ was my estimate at the time. This was with Fireworks (ZDR) which has since increased their prices (and got slower!).

That being said, very impressed with the model, and looking forward to what comes next. As the frontier models become less subsidized, the open models will become more appealing.

P.S. There are subscription plans for open models, but I've found most of them to be extremely slow, have model throttling (only so much of model X), and also very sketchy about training and data retention. No thanks! If you want to share your data, just use Muse Spark contributor. Seems impossible to beat that on price per task if you don't mind feeding your data to the Meta machine (spoiler: I won't).

Who are these "people"? I'm certainly not one of them, and I consider myself a progressive.

In an ideal society, I think you just tax digital ads to help compensate for the damage from this attention economy. Maybe even use that money for education?

Look up Section 230 of the Communications Decency Act of 1996. It explains the current situation and should be obvious that it's less than perfect. Basically without this then online platforms would be responsible for the content they host (publishers) and would have been sued into oblivion before any of them got up and running... especially Facebook with it's beginnings. So to compensate there's the good samaritan provision which is the whole censorship thing from the right's perspective (like back during Covid) and the protection from being a publisher let's them host hate speech from the left's perspective.

Personally I think making them public square is probably the only real solution if you want to keep social media and avoid this drama, or make them liable as publishers and these platforms will disappear overnight (maybe the best result?!?)...

This carries over into LLMs I presume, which have bias and I assume are not being sued for the content the publish? I'm not really sure about how it applies to LLMs to be honest.

Regardless, other countries will have different laws, that have different takes on censorship and what private companies are responsible for.

I was gonna say if you want to get away with a little hacking, now would be a great time to spoof ChatGPT.

> See what they did at Volkswagen scamming the system and then refusing to own up to it.

If you mean Dieselgate, then agreed, horrible. It is worth noting that after the scandal we found out that a whole bunch of (non-german like Suzuki and Hyundai) manufacturers were actually worse (NOx) with emissions well beyond legal limits. It just didn't make the headlines nearly as much.

So while we can absolutely criticize VW, I don't think it's fair to generalize this as a german issue.

I did this yesterday with user data to find duplicates and determine which conflicting fields should win in a merge. Took me about an hour-ish to run the experiment. I provided a good amount of context for each pair. It was underwhelming / OK at best and didn't add any value to my existing merge workflow, so I threw it out after the experiment (I kept the findings though). It was quite easy to setup and use via openrouter and extremely cheap and fast.

For fun I also tried using it at my local LLM router (all of my prompts and responses go through it for personal analytics) to decide which model to route tasks to based complexity etc. Again, underwhelming for my purposes, and so I'm not using it.

Not sure what real world use case it's best at, but I agree that it's simple enough to implement it yourself and see if it fits the type of work you are doing.

Yes, I'm still supporting two large-ish rails applications and a larger Rust application.

Yeah. I'm going through an exercise right now with a large codebase where I'm split some of it up into crates so that iterating on specific parts incurs less build overhead. Literally changing my architecture... which annoyed me a bit, but it does make sense.

Rust for sure. Builds become a problem eventually though. Elixir is much more interesting now that they added types (I think they're doing more with this still?). I quite like Gleam from a human (non-agent) perspective.

> Rails is so into agentic coding that they publish benchmarks of how effective different models are for building with rails

To each their own, but I think using Rails in the era of agents is probably a mistake. It has 2 big problems IMO.

(a) A ton of footguns that an agent can and will miss on occasion, and as we manually review less and less code, we will also miss it, and

(b) security on the web has become increasingly important due to hostile AI agents, and while rails has great secure conventions (see (a)), it's still easier to screw up something and not catch it until after it's deployed because there are simply less guardrails in place (dynamic types, no compile, slow tests, generally bad performance profile).

Types are great. No nulls are great. Exhaustive switches are great. Derive macros and Serde are great. Detailed errors are great.

However Rust has one serious problem, BUILDS. It’s slow and it takes up a ton of space and memory. You will be working around this on any sizeable codebase. Using cargo check to ration your compiles, a dedicated build box, etc. Now imagine you're using it for a web application with compiled maud templates... you're potentially waiting minutes to see your changes.

If we accept that over half of the people building with rust are using agents (I'm sure this is much higher in reality), then you should also accept that there is a ton of time and resources spent on building / compiling. Optimizing this should probably be a top priority. They need someone who will be fanatical about shaving time and memory use from the build process.

> t's a bad idea to pay someone to remove a problem they created for you.

Worth noting that this is what most of the live service gaming industry is doing.

> I think the more comparable thing would be youtube premium (paying to not have as many ads there), and I certainly pay for that.

This isn't comparable, because the base product is free. We're talking about paying an upfront fee for a premium product (like $2k for a phone) and then being shown ads. Youtube would be a vastly different product if everyone had to pay a one time $2k to use it (and then purchase premium to unlock ad-free). It's also noticeable that a lot of the "premium" media services go from being ad-free to ad-light... so we're back to cable... but that's a different discussion.

Maybe. How are the seat heater subscription services going for BMW? I thought the market had clearly rejected it, but maybe they powered through and people have begrudgingly accepted it now?

I don't think people want to pay a large sticker price and then an additional monthly fee on top of it to enable features or disable ads... but I honestly don't know. I just know I sure af won't lol.

In general, no where yet. I do think the smart phone interface (iOS) could become at least somewhat obsolete soon. As we move more towards a primarily inference interface it will open up the field to more competitors (less moat). That's just my opinion though. Be interesting, as usual, to see what comes out of China in the next few years.

Imagine spending $2k on a phone only to be presented with ads. Quick way to lose your luxury and loyal customers IMO.

On the flip side, imagine having another product made for people who can afford $2k phones... Well you'd probably pay a lot to advertise to that market.

I think trying to balance between these 2 is going to be tough, and your luxury market could evaporate if a significant competitor came along. And by competitor, we're probably just talking about AI in your pocket or on your face that replaces the traditional phone interface. I also think it's a real slippery slope to short term profits once you start taking ad revenue.

I would agree, but I'm not sure I trust the "no training" policy. For DS 4.1 Flash I was able to use a US ZDR provider.

Murica just has a MAGA problem. We can still be friends if and when you sort that out. Us Canadians like most of you quite a lot.

API rates still aren’t quite competitive with the OpenAI x20 accounts, but they are definitely getting close with deepseek 4.1 flash. I spent a few days with only 4.1 and was very impressed.

So I actually struggled with this recently (2 months ago). My "vision" was to have customers BYO agent and provide a secure MCP with an in-app (forced) approval gate for sensitive operations. But everytime I described to one of my clients, how they could bring their own ChatGPT, Claude, Gemini, w/e, I got blank stares.

So now I'm building (already have done so mostly) cheap ZDR models into the application. I'm trying to build them in a way that I would want to use them, so a less "I'll be your AI today!" help bubble, but who's kidding... At least they aren't just customer pacifiers that use the help system, but actually can act on behalf of the user.

After which I'll end up building the MCP, but for power users, hopefully leveraging much of the same work. It's worth noting that MCPs have way too much friction still. It's gotten a lot easier recently (very recently) to bring your own custom MCP with Claude, but ChatGPT is more work, and restrictive (specific account types or submitted MCP apps) for the end user and I don't even know what Google wants at this point... It's a bit messy. IMO they are all dropping the ball (except Claude, which has taken the best approach).

I think we might be in this interim state that requires us to build ALL of the UIs, to address ALL of the users (Human responsively and Clankers), and it's a bit painful. If I was "the user" myself, then I would just want a CLI w/ OAUTH with clear documentation, that an agent could use (how is this worse than MCP?). But I'm not my own customer. That being said... I probably WILL build this version to scratch my own itch and bet on the future.

> if it doesn't want to leave.

This is where the lawyers find the loophole to get the cat out of the bag.

Prompt fatigue with nukes. --dangerously-skip-permissions

> I believe people are getting sick of every interaction being monetized and are finally seeking out spaces where you can learn and interact with others without that threat.

Shhh... you're going to get libraries monetized.

Luna is great but makes a lot of mistakes at high and lower in my experience (large rust codebase). I use Luna Max for asynchronous subagent reviews and am very happy with its work, but it’s slow af.