Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models • Hacker News Board

Re: Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models

#251

Post by dang » Tue, Jul 28, 2026, 2:56 AM UTC

leonkatz wrote: post #249

dang wrote: post #242

Earlier quoted context omitted.

I know it always feels like the mods are against you and secretly in cahoots with the other side when you get a moderation reply like the GP, which obviously doesn't feel good. https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que... But I promise you this is not so. We've warned and banned countless accounts on both sides of every divisive political question, including ideological ones and nationalistic ones.…

dang - sorry to flag this here, but I've emailed hn@ycombinator.com twice over about 11 days with no reply, and I'm not sure it's reaching you. My account (leonkatz) seems to be in a shadowbanned state - new comments show up [dead] on arrival (e.g. a genuine reply I left on the "Little Book of Reinforcement Learning" thread). I'm a real person, not a spammer - I'm building an open-source project (github.com/rekol-io/…

I'm sorry we didn't reply to your emails - it's because we're inundated with so many emails that we can no longer even look at them all, let alone respond to them all. But since it may be of interest to readers, I'll share here what I would have sent to you in a reply:

Your posts are getting killed because our software classified the text as genai. That's not allowed on HN - see https://news.ycombinator.com/newsguidelines.html#generated and https://news.ycombinator.com/item?id=47340079.

Can you write by hand any text that you plan to post to HN? Here's an important tip we send to users about this:

Write any text that you post to HN by hand. Don't use an LLM to generate any of it (not even a tiny bit, including to edit or spruce it up). Reason: the community is super fussy about this right now, and LLM language has a certain quality that is generating quite some backlash when it appears on HN itself. This is a big dividing line at present!

Re: Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models

#252

Post by leonkatz » Tue, Jul 28, 2026, 4:32 AM UTC

dang wrote: post #251

leonkatz wrote: post #249

Earlier quoted context omitted.

dang - sorry to flag this here, but I've emailed hn@ycombinator.com twice over about 11 days with no reply, and I'm not sure it's reaching you. My account (leonkatz) seems to be in a shadowbanned state - new comments show up [dead] on arrival (e.g. a genuine reply I left on the "Little Book of Reinforcement Learning" thread). I'm a real person, not a spammer - I'm building an open-source project (github.com/rekol-io/…

I'm sorry we didn't reply to your emails - it's because we're inundated with so many emails that we can no longer even look at them all, let alone respond to them all. But since it may be of interest to readers, I'll share here what I would have sent to you in a reply: Your posts are getting killed because our software classified the text as genai. That's not allowed on HN - see https://news.ycombinator.com/newsguide…

Thank you for responding and for allowing me back from purgatory. I didn't realize there was a ban on GenAI. I have used it to correct my posts. They do a pretty good job of cleaning up my slop. Thanks again and I won't make the same mistake in the future.

Re: Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models

#253

Post by jambalaya8 » Thu, Jul 30, 2026, 10:18 PM UTC

mattjoyce wrote: post #216

jambalaya8 wrote: post #150

Earlier quoted context omitted.

I am too young (most of us on here are) to have lived through the paying for time on time-share machines in the 60s/70s, but this is giving me creepy memories of paying for sprintnet/telenet and tymnet... And I guess aol, compuserv, delphi. Are we really doing this computing model again?

well spotted, I couldn't place the feeling, me it's like paying per minute phone costs when downloading on a modem. You know it will cost, but you do it anyway. The other weird thing about the subs is that if the agents aren't grinding if feels like I'm losing money.

I still remember searching really hard for shareware apps and games between 1 to 3 megs, though I am sure I used more minutes looking for them than the downloads would have taken. Back then, if we weren't limited by our dialups, we were surely limited by our very low storage capabilities. But man, code was so much better and tighter then. Haven't seen much focus on this in vibe coding, sadly.

Re: Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models

#254

Post by springtimesun » Fri, Jul 31, 2026, 11:51 AM UTC

XCSme wrote: post #149

springtimesun wrote: post #124

I have been trying something like this with local only models. I think I’ve tried everything that will fit in 96GB alone and in concert with each other using real project data (Rails). I haven’t found much benefit to anything other than Qwen 3.6 27b dense feeding a Claude final pass. I really thought layering was going to work like the law of large numbers, but for my workload it just hasn’t.

I have a spare 3090 that I want to use to off-load some tasks from Claude to a local model (probably Qwen 3.6 27b), any success with that? Is it good enough to follow some tasks, coding requirements or browser usage?

Alone, Qwen can close basic bug tickets. Passing up and passing down have not really worked that well for me. In some cases passing Qwen’s output up to Claude can be actively harmful bc it can anchor the solution space to a bad approach. Passing down isn’t worth the time. By the time I have Claude plan something, most of the time it’s 1/3 or 1/4 more tokens to just have Claude do the implementation.

Where it has worked is to have Qwen not just write the proposed solution, but also include all the files it consulted and why it did or didn’t consider them. The documentation is often longer than the patch, but it lets me prompt Claude differently. I tell Claude to challenge the implementation and find holes, edge cases and things not considered. Maybe it’s my imagination, but Claude kicks it into another gear when you tell it to evaluate the results of another model.

Maybe this is a pure preference but I find Qwen to be bad at UI and these edits Claude will often let go.

One thing I haven’t tried, but will at some point is to hand Qwen all my half assed ideas and things I can’t be bothered to spec out. Probably not to take the output, but to help me draft a more complete issue because it’s only costing me electricity.