There's a blog post in this for sure. And a classic joke.","title":null,"type":"comment","url":null},{"author":"victorhooi","children":[],"created_at":"2026-11-09T23:53:33.000Z","created_at_i":1794268413,"id":50434118,"options":[],"parent_id":50433063,"points":null,"story_id":50429024,"text":"Sorry I have to ask....um....sniffing class? What is that?
Btw, loved your work with Django!","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T22:17:03.000Z","created_at_i":1794262623,"id":50433063,"options":[],"parent_id":50429659,"points":null,"story_id":50429024,"text":"I was at my dog's sniffing class for this one, but I had Codex on my Mac (controlled from Codex on my iPhone) run the test prompt with my `llm-meta-ai` plugin, which has the ability to refresh to fetch new models that Meta AI have released: https://github.com/simonw/llm-meta-ai#usage","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:40:23.000Z","created_at_i":1794249623,"id":50429659,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Incredible. How do you get these out so fast?","title":null,"type":"comment","url":null},{"author":"permalac","children":[{"author":"MadameMinty","children":[],"created_at":"2026-11-09T18:58:28.000Z","created_at_i":1794250708,"id":50429943,"options":[],"parent_id":50429664,"points":null,"story_id":50429024,"text":"I like medium best, but minimal is not bad at all.","title":null,"type":"comment","url":null},{"author":"mring33621","children":[],"created_at":"2026-11-09T20:06:44.000Z","created_at_i":1794254804,"id":50431170,"options":[],"parent_id":50429664,"points":null,"story_id":50429024,"text":"Yeah minimal has style","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:40:53.000Z","created_at_i":1794249653,"id":50429664,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Is there a problem if I like the minimal more than the others?","title":null,"type":"comment","url":null},{"author":"imimayj1337","children":[{"author":"swatcoder","children":[{"author":"flir","children":[],"created_at":"2026-11-09T19:12:43.000Z","created_at_i":1794251563,"id":50430186,"options":[],"parent_id":50429911,"points":null,"story_id":50429024,"text":"You could use a second prompt to z-order the objects (I tend to break down tasks and pipeline them, rather than trying to stuff everything in a one-shot prompt. Seems to work better for me, could be a holdover from the early days).","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:56:38.000Z","created_at_i":1794250598,"id":50429911,"options":[],"parent_id":50429670,"points":null,"story_id":50429024,"text":"They're not. It's an illusion from the composition.
If you look at the code for each of these (and probably most across models), you can see that the drawing elements are organized by concept. Presumably, this helps accuracy by keeping attention on one concept at a time, but you end up with a stack of conceptualized objects and lose the ability to interleave things.
No doubt these could all do better with a more thorough and specific prompt that coached them away from that technique, but otherwise it's a very subtle detail that a "thinking" process needs to stumble upon independently.","title":null,"type":"comment","url":null},{"author":"latexr","children":[],"created_at":"2026-11-09T19:03:42.000Z","created_at_i":1794251022,"id":50430039,"options":[],"parent_id":50429670,"points":null,"story_id":50429024,"text":"They\u2019re not. Note how both pedals are on the same side of the frame. That\u2019s wrong!","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:41:18.000Z","created_at_i":1794249678,"id":50429670,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"It does seem pretty insane to me that only on max setting (and after waiting 27 seconds) did you actually manage to receive an SVG where it appears that the pelican's legs are straddling the bike correctly.
Edited to make clear what it looks like to me and what I was looking for.","title":null,"type":"comment","url":null},{"author":"eterm","children":[],"created_at":"2026-11-09T18:52:25.000Z","created_at_i":1794250345,"id":50429837,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Did minimal just hand it off to something else? It seems completely out of line with the rest.","title":null,"type":"comment","url":null},{"author":"jolt42","children":[],"created_at":"2026-11-09T18:54:47.000Z","created_at_i":1794250487,"id":50429873,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Pelican is the new teapot","title":null,"type":"comment","url":null},{"author":"djmips","children":[],"created_at":"2026-11-09T19:15:04.000Z","created_at_i":1794251704,"id":50430237,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"birds at minimal and return at max","title":null,"type":"comment","url":null},{"author":"phyrex","children":[{"author":"santoshalper","children":[],"created_at":"2026-11-09T21:20:27.000Z","created_at_i":1794259227,"id":50432338,"options":[],"parent_id":50430245,"points":null,"story_id":50429024,"text":"Wearing a helmet is smart, so that checks out.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:15:20.000Z","created_at_i":1794251720,"id":50430245,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"I like that from xhigh on up its wearing a helmet.","title":null,"type":"comment","url":null},{"author":"small_model","children":[],"created_at":"2026-11-09T19:32:12.000Z","created_at_i":1794252732,"id":50430556,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Not sure this is a good way to benchmark LLM's, no offence, but a single SVG is hardly a measure of a models capability.","title":null,"type":"comment","url":null},{"author":"AlecSchueler","children":[{"author":"swatcoder","children":[{"author":"mvkel","children":[],"created_at":"2026-11-09T20:10:01.000Z","created_at_i":1794255001,"id":50431223,"options":[],"parent_id":50430884,"points":null,"story_id":50429024,"text":"This is interesting. There's a balance between:
- simple prompt for complex output (bad)
- complex prompt for simple output (bad)
- simple prompt for simple output (good)
- complex prompt for complex output (.... is this good or bad?)","title":null,"type":"comment","url":null},{"author":"viraptor","children":[],"created_at":"2026-11-09T21:01:25.000Z","created_at_i":1794258085,"id":50432067,"options":[],"parent_id":50430884,"points":null,"story_id":50429024,"text":"To be fair, people have a genuine preference for images with things facing to the right. And lots of drawings will do 3 motion lines. You'd likely draw a dashed rather than solid separator too, or it would look like a random line. Lots of the AI habits are there because of choices we make.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:49:02.000Z","created_at_i":1794253742,"id":50430884,"options":[],"parent_id":50430682,"points":null,"story_id":50429024,"text":"Same reason they're all facing right with the same composition, they all have three motion lines, they all have a yellow sun and 2-3 blobby clouds, they're all on two-lane pavement with a white dashed lane separator.
And the same reason Claude has all its notorious Claudisms in prose.
The shorter and vaguer a prompt you give to an LLM, the more it fills in further details using associations they've developed through training. These are ley lines that the models gravitate towards. Diffusion models too.
It's why lazy AI use for creative work always comes across as lazy in the wild and always will.
When you use AI for "creative" purposes yourself, it feels like magic to see a pelican riding a bicycle at all, and you're impressed by all those cool details it added. But then 10, or 100, or 10M people throw more or less the same trivial prompts at it and we start to recognize all those "ley line" associations fill in the gaps the same way again and again and again.","title":null,"type":"comment","url":null},{"author":"Buttons840","children":[],"created_at":"2026-11-09T22:16:25.000Z","created_at_i":1794262585,"id":50433054,"options":[],"parent_id":50430682,"points":null,"story_id":50429024,"text":"There's only one average in a set.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:38:00.000Z","created_at_i":1794253080,"id":50430682,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Any idea why there's so much convergence on the fish in a basket idea?","title":null,"type":"comment","url":null},{"author":"ezst","children":[{"author":"ourcat","children":[{"author":"9dev","children":[],"created_at":"2026-11-09T22:57:13.000Z","created_at_i":1794265033,"id":50433519,"options":[],"parent_id":50432672,"points":null,"story_id":50429024,"text":"At this point, even if the labs wouldn\u2019t actively pelican-max their models, isn\u2019t the training corpus full of pelicans already? Funny to think how one models choice of adding a red scarf in one of Simon\u2019s earlier experiments might have led to an archetype of pelicans riding bicycles\u2026","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T21:43:23.000Z","created_at_i":1794260603,"id":50432672,"options":[],"parent_id":50430710,"points":null,"story_id":50429024,"text":"Indeed. And that they were all trained to give it a red scarf, which was not asked for.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:38:58.000Z","created_at_i":1794253138,"id":50430710,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Wouldn't that be a sign that those models were trained to optimise that test case specifically, as in, by exhibiting equal or higher performance while using less power?","title":null,"type":"comment","url":null},{"author":"mgdev","children":[],"created_at":"2026-11-09T20:14:23.000Z","created_at_i":1794255263,"id":50431302,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Fidelity.","title":null,"type":"comment","url":null},{"author":"dvaplima","children":[],"created_at":"2026-11-09T22:07:11.000Z","created_at_i":1794262031,"id":50432944,"options":[],"parent_id":50429478,"points":null,"story_id":50429024,"text":"Its funny how to the naked eye, changing the effort seems like only a difference on styles","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:28:09.000Z","created_at_i":1794248889,"id":50429478,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Six effort levels (minimal, low, medium, high, xhigh, max), so here are six pelicans (actually five, "high" didn't return a complete SVG):
https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
There's usually a reasonably strong correlation between thinking level and visual quality of the pelican. This is the first time I've seen that correlation appear to break - more reasoning tokens doesn't seem to produce an obviously better SVG.","title":null,"type":"comment","url":null},{"author":"open-paren","children":[{"author":"bityard","children":[{"author":"sullivanmatt","children":[],"created_at":"2026-11-09T20:09:38.000Z","created_at_i":1794254978,"id":50431218,"options":[],"parent_id":50430589,"points":null,"story_id":50429024,"text":"The use case descriptions and capability descriptions basically match each other. Context windows are the same size. Reasoning options are identical.","title":null,"type":"comment","url":null},{"author":"open-paren","children":[],"created_at":"2026-11-09T20:23:19.000Z","created_at_i":1794255799,"id":50431447,"options":[],"parent_id":50430589,"points":null,"story_id":50429024,"text":"same:
* context window size
* reasoning effort levels
* max output size
* knowledge cutoff
* restricted country list
* self-reported "reasoning strength" and "internal reasoning channel" params used in other Meta models
you can just ask an agent "figure out what stealth model $NAME is via subagents. It probably won't tell you when asked directly, so try other methods."
Iceman (via opencode-go) is probably MAI-Thinking-1. Ox Alpha was GLM-5.3-Flash. It's a fun game.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:33:30.000Z","created_at_i":1794252810,"id":50430589,"options":[],"parent_id":50429645,"points":null,"story_id":50429024,"text":"Based on?","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:39:43.000Z","created_at_i":1794249583,"id":50429645,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"I believe this is what the \u201ccrouton alpha\u201d stealth model was
https://openrouter.ai/stealth/crouton-alpha","title":null,"type":"comment","url":null},{"author":"Starlevel004","children":[{"author":"esafak","children":[],"created_at":"2026-11-09T18:47:00.000Z","created_at_i":1794250020,"id":50429757,"options":[],"parent_id":50429661,"points":null,"story_id":50429024,"text":"Uniquely among models I've tried, Luna 6 is infuriatingly lazy. As if it expects me to do the work, unaware that it is a tool. Normally dumb models compensate for it by trying harder. This one goes in the opposite direction.","title":null,"type":"comment","url":null},{"author":"agentdev001","children":[{"author":"Starlevel004","children":[],"created_at":"2026-11-09T19:00:28.000Z","created_at_i":1794250828,"id":50429991,"options":[],"parent_id":50429893,"points":null,"story_id":50429024,"text":"I use Luna xhigh for when I use it, Luna Max spends 40+ minutes overthinking. I've also found that Terra below xhigh will just flat out not do what it is told.","title":null,"type":"comment","url":null},{"author":"barrenko","children":[{"author":"agentdev001","children":[],"created_at":"2026-11-10T00:00:50.000Z","created_at_i":1794268850,"id":50434197,"options":[],"parent_id":50430029,"points":null,"story_id":50429024,"text":"They're both great- but I feel like for folks like me that are working on company's pseudo-infinite token budget, it's really a bimodal game. Im either reaching for the best cheapest for fanned out benchmark runs, or the fastest best for everything else. 6.1 Sol ultrafast is the latter for me, so Terra doesn't really compete with it. Terra has certainly replaced 6 Luna for me, however.
(And yes Astra is a thing, but if im considering Astra, im usually opting for an Anthropic model, 90% of the time.)","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:02:51.000Z","created_at_i":1794250971,"id":50430029,"options":[],"parent_id":50429893,"points":null,"story_id":50429024,"text":"What is your feel re. Terra vs. Sol?","title":null,"type":"comment","url":null},{"author":"theanonymousone","children":[{"author":"tryptophan","children":[],"created_at":"2026-11-09T22:02:06.000Z","created_at_i":1794261726,"id":50432887,"options":[],"parent_id":50431034,"points":null,"story_id":50429024,"text":"My experience was the same. It is practically five to ten times more expensive than deepseek or luna.
The artificial analysis graph is not very accurate.","title":null,"type":"comment","url":null},{"author":"agentdev001","children":[],"created_at":"2026-11-09T23:54:01.000Z","created_at_i":1794268441,"id":50434119,"options":[],"parent_id":50431034,"points":null,"story_id":50429024,"text":"Im surprised, when I've worked through benchmarking things I've run into comparable costs for Luna Max vs Terra Low. Also anecdotal from me, of course.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:57:49.000Z","created_at_i":1794254269,"id":50431034,"options":[],"parent_id":50429893,"points":null,"story_id":50429024,"text":"> Costs seem similar
They weren't, unfortunately, every time I tried. Just an anecdote, of course.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:55:52.000Z","created_at_i":1794250552,"id":50429893,"options":[],"parent_id":50429661,"points":null,"story_id":50429024,"text":"Give Terra 6.1 Low a spin in places where you'd have previously used Luna Max. Costs seem similar, due to Terra being much smarter: thus requiring far fewer tokens per task.
Looking solely at AA's intelligence index, recognizing the pitfalls with using that as a measure- Luna 6 Max > Terra 6.1 Low is 51.3% cheaper, 94.1% less output tokens, for 1.56 higher index score.","title":null,"type":"comment","url":null},{"author":"all2","children":[],"created_at":"2026-11-09T19:23:16.000Z","created_at_i":1794252196,"id":50430383,"options":[],"parent_id":50429661,"points":null,"story_id":50429024,"text":"Tbf, I would take Luna 6 baked to silicon. It is that good. I would also take kimi k3 baked to silicon.
These are both models that I could see using on a long horizon.","title":null,"type":"comment","url":null},{"author":"flir","children":[],"created_at":"2026-11-09T19:25:55.000Z","created_at_i":1794252355,"id":50430438,"options":[],"parent_id":50429661,"points":null,"story_id":50429024,"text":"> Imagine saying this about anything else
Javascript frameworks, circa 2016 ;)","title":null,"type":"comment","url":null},{"author":"theanonymousone","children":[],"created_at":"2026-11-09T19:56:37.000Z","created_at_i":1794254197,"id":50431007,"options":[],"parent_id":50429661,"points":null,"story_id":50429024,"text":"> I've been hoping for Luna 6.1/2 as the next model release for a few days now.
Same. It became my workhorse after the 5.6 price drop. Still so, although it's showing its age, as you said.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:40:36.000Z","created_at_i":1794249636,"id":50429661,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Luna 6 is getting a bit long in the teeth now. It's two full months old, but it was only a slight step above Luna 5.6, which is already four months old (Imagine saying this about anything else). So I'm not too surprised that a newer small model is beating it handily. I've been hoping for Luna 6.1/2 as the next model release for a few days now.","title":null,"type":"comment","url":null},{"author":"caaqil","children":[{"author":"esafak","children":[],"created_at":"2026-11-09T18:44:28.000Z","created_at_i":1794249868,"id":50429717,"options":[],"parent_id":50429683,"points":null,"story_id":50429024,"text":"They are catching up; the Spark series is good.","title":null,"type":"comment","url":null},{"author":"achandra03","children":[],"created_at":"2026-11-09T18:46:32.000Z","created_at_i":1794249992,"id":50429748,"options":[],"parent_id":50429683,"points":null,"story_id":50429024,"text":"Isn't Muse Wonder at or at least very close to the frontier? Considering they essentially started from scratch after rebooting their AI efforts post-Llama 4 about a year ago it's pretty solid progress.","title":null,"type":"comment","url":null},{"author":"StarlessHours","children":[],"created_at":"2026-11-09T18:47:00.000Z","created_at_i":1794250020,"id":50429756,"options":[],"parent_id":50429683,"points":null,"story_id":50429024,"text":"Meta is trying to stay relevant. They fell behind after llama 4, hired new talent and re-organized.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T18:41:57.000Z","created_at_i":1794249717,"id":50429683,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"I'm always a bit surprised to see Meta talking about models. They have become (almost*) entirely irrelevant to the LLM race, which is weird because their AI lab used to be decent before LLM was cool.
What happened to zuck? Was the metaverse this destructive?
(*) Feeling charitable today","title":null,"type":"comment","url":null},{"author":"dom96","children":[],"created_at":"2026-11-09T18:46:47.000Z","created_at_i":1794250007,"id":50429751,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Super frustrating that I still can\u2019t give Muse Wonder Full a spin.","title":null,"type":"comment","url":null},{"author":"gopalv","children":[],"created_at":"2026-11-09T18:46:55.000Z","created_at_i":1794250015,"id":50429752,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Model refusals should be a -ve score, not zero?
Treating the refusal as a 0 score erases the difference between lack of capability and a lack of willingness to perform.","title":null,"type":"comment","url":null},{"author":"Pragmata","children":[],"created_at":"2026-11-09T18:50:55.000Z","created_at_i":1794250255,"id":50429811,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"No weights? Shame","title":null,"type":"comment","url":null},{"author":"aaroninsf","children":[{"author":"anderber","children":[{"author":"herbst","children":[{"author":"anderber","children":[],"created_at":"2026-11-09T20:10:57.000Z","created_at_i":1794255057,"id":50431235,"options":[],"parent_id":50431204,"points":null,"story_id":50429024,"text":"More than SpaceX? Or OpenAI and it's agreement with military weapons? Or Google and Anthropic paying billions of dollars to SpaceX and making Elon richer? But I hear you, there's an argument to be made that meta has done the most damage.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T20:08:47.000Z","created_at_i":1794254927,"id":50431204,"options":[],"parent_id":50430477,"points":null,"story_id":50429024,"text":"While this is true meta is argueable making the world worse the longest, probably did more damage in that time too.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:28:03.000Z","created_at_i":1794252483,"id":50430477,"options":[],"parent_id":50430128,"points":null,"story_id":50429024,"text":"I honestly have a hard time trying to put these AI companies in some sort of moral hierarchy. They seem to all have issues... I feel like at least with open models you can have more ethical services.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:09:11.000Z","created_at_i":1794251351,"id":50430128,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Never Meta.
Seriously. Moral harm is real and there is no excuse.","title":null,"type":"comment","url":null},{"author":"BatchJob","children":[{"author":"40four","children":[],"created_at":"2026-11-09T19:22:50.000Z","created_at_i":1794252170,"id":50430376,"options":[],"parent_id":50430144,"points":null,"story_id":50429024,"text":"https://www.thebiglebow.ski/our-most-modestly-priced-recepta...","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:10:04.000Z","created_at_i":1794251404,"id":50430144,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"it is our most modestly priced receptacle...","title":null,"type":"comment","url":null},{"author":"ecshafer","children":[{"author":"40four","children":[{"author":"nicce","children":[{"author":"ac29","children":[],"created_at":"2026-11-09T21:46:44.000Z","created_at_i":1794260804,"id":50432713,"options":[],"parent_id":50431427,"points":null,"story_id":50429024,"text":"Muse Spark Open is 816B-A42B: https://huggingface.co/meta-models","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T20:21:48.000Z","created_at_i":1794255708,"id":50431427,"options":[],"parent_id":50430645,"points":null,"story_id":50429024,"text":"Tried to use the download button from the page, but it failed miserable. First, could not press the button because had to decline cookies first. Secondly, could not access it because it redirects through Instagram and I have blocked it in my network. So I still don't know how big it is...","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:36:22.000Z","created_at_i":1794252982,"id":50430645,"options":[],"parent_id":50430153,"points":null,"story_id":50429024,"text":"They just released Muse Spark Open for that","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:10:39.000Z","created_at_i":1794251439,"id":50430153,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"I would love it if Meta would release these models to use locally to build up some usage while they are lagging.","title":null,"type":"comment","url":null},{"author":"kbrannigan","children":[{"author":"grim_io","children":[],"created_at":"2026-11-09T19:37:00.000Z","created_at_i":1794253020,"id":50430659,"options":[],"parent_id":50430375,"points":null,"story_id":50429024,"text":"I don't think so.
Cross discipline skills are more valuable in the real world, outside of hyper specialization seen in the California tech scene. Even there, it's coming to an end.","title":null,"type":"comment","url":null},{"author":"drillsteps5","children":[{"author":"ElectricalUnion","children":[],"created_at":"2026-11-09T23:42:57.000Z","created_at_i":1794267777,"id":50433998,"options":[],"parent_id":50432839,"points":null,"story_id":50429024,"text":"Unless your task happens to be completely contained in the set of your "curated works", that is, trivial and computer-sciency, the resulting model is likely to have for example absolutely no concept of common sense, society, the natural world, what humans consider artistic or of good taste. It will probrably make, say, a good and correct compiler, but fail spetacularly on anything else more complex that a toy crud.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T21:57:57.000Z","created_at_i":1794261477,"id":50432839,"options":[],"parent_id":50430375,"points":null,"story_id":50429024,"text":"Exactly what I was thinking. If this model will be used for coding, does it need to be fed the entire (mostly pirated) collection of works from Anna's Archive? Probably not. Maybe some, primarily computer-science oriented.
The main corpus would be a collection of thoroughly reviewed, clean, logical, well-documented, and secure code in whatever languages this thing will need to be an expert in. (Edit: also likely some collection of Q&A, stackoverflow-slyle).
This way we could also look into reasonable compensation agreement with authors of the code and the books/articles, if and when applicable.
Has no one tried that?","title":null,"type":"comment","url":null},{"author":"GuB-42","children":[],"created_at":"2026-11-09T23:26:42.000Z","created_at_i":1794266802,"id":50433841,"options":[],"parent_id":50430375,"points":null,"story_id":50429024,"text":"Human lawyers, coders, accountants and designers are a lot more than their jobs.
They are people with common knowledge, a lawyer has to understand carpentry to understand how to deal with how a client's roof fell off, a coder has to understand transportation for writing a travel app, an accountant has to understand medicine to do accounting for a pharmacy, a designer has to understand cooking to design a toaster. Not expert knowledge, but enough to understand that toasters are hot and need electricity.
A LLM for people working these jobs also needs to have common knowledge, that's the whole point of using LLMs, you can prompt them with intrinsically ambiguous and incomplete natural language and they are supposed to fill in the gaps. And the more common knowledge they have, the better they are at it. So why settle for less? If it is an optimization problem, we already have the solution, it is the "mixture of experts", which is better at optimizing then having an "lawyer LLM", "coder LLM", etc... because it is better adapted to the internal structure of the neural network behind it.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:22:45.000Z","created_at_i":1794252165,"id":50430375,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"This is starting to feel just like the early PC days. A bunch of Computer models coming out, all shapes and color, but eventually it settled around mouse keyboard, monitor
most people, use the office suite: Word, Spreadsheet , presentations, database.
I bet we're going to see the exact same thing with LLMs at specialized models instead of the everything models.
One for lawyers, coders, Accountants, Designers, etc\u2026","title":null,"type":"comment","url":null},{"author":"slopnt","children":[],"created_at":"2026-11-09T19:23:26.000Z","created_at_i":1794252206,"id":50430386,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"I use haiku and luna for support on most my SWE tasks since they are cheap and fast. I find them enough for most things if you are patient, and the fact an even cheaper model for comparable intelligence is out there now is great news!","title":null,"type":"comment","url":null},{"author":"johnnyApplePRNG","children":[],"created_at":"2026-11-09T19:26:13.000Z","created_at_i":1794252373,"id":50430446,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Good for summarizing your tweets or what?! Is this a model for ants?!","title":null,"type":"comment","url":null},{"author":"mococa","children":[{"author":"dvaplima","children":[],"created_at":"2026-11-09T22:12:09.000Z","created_at_i":1794262329,"id":50433000,"options":[],"parent_id":50430458,"points":null,"story_id":50429024,"text":"Usually wait for 6mo, but the gap is getting narrower
https://ai2027-tracker.com/predictions/china-model-gap/","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:27:09.000Z","created_at_i":1794252429,"id":50430458,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Not open weight? Meh... I'll wait for the Chinese releases :)","title":null,"type":"comment","url":null},{"author":"v9v","children":[{"author":"system2","children":[],"created_at":"2026-11-09T21:46:58.000Z","created_at_i":1794260818,"id":50432714,"options":[],"parent_id":50430725,"points":null,"story_id":50429024,"text":"American SOTA still cannot compete with Chinese models. They will always pretend there is no elephant in the room.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:39:54.000Z","created_at_i":1794253194,"id":50430725,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Why no comparisons with DeepSeek?","title":null,"type":"comment","url":null},{"author":"minimaxir","children":[{"author":"system2","children":[],"created_at":"2026-11-09T21:45:25.000Z","created_at_i":1794260725,"id":50432691,"options":[],"parent_id":50430835,"points":null,"story_id":50429024,"text":"No batch. Pass.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T19:46:12.000Z","created_at_i":1794253572,"id":50430835,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"> It costs $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens,
Notable cache cost of 4% input, compared to Luna/Haiku's 10% of input.","title":null,"type":"comment","url":null},{"author":"segmondy","children":[],"created_at":"2026-11-09T19:49:34.000Z","created_at_i":1794253774,"id":50430893,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"What makes it small? Anyone know the size? I don't see it anywhere. Small in relation to their current line up? Anyways, last I tried Meta was llama4. I stepped aside not because of how bad llama4 got, but also because of Zuck's move. Anyone here using Meta? I don't know anyone in my circle using it, but I'm sure folks are using it.","title":null,"type":"comment","url":null},{"author":"cheesecompiler","children":[{"author":"Pedro_Ribeiro","children":[{"author":"mgdev","children":[],"created_at":"2026-11-09T20:15:27.000Z","created_at_i":1794255327,"id":50431319,"options":[],"parent_id":50431301,"points":null,"story_id":50429024,"text":"Please. Complaints are the last thing that we have humans have.","title":null,"type":"comment","url":null},{"author":"cheesecompiler","children":[{"author":"groggo","children":[],"created_at":"2026-11-09T21:10:34.000Z","created_at_i":1794258634,"id":50432203,"options":[],"parent_id":50432120,"points":null,"story_id":50429024,"text":""Government announces AI ban to avoid further cheesecompiler burnout".
haha just kidding. I agree. That's what benchmarks are for. Just keeping a rough understanding of what companies have which models, and understanding general trends of capabilities, economics, and performance of models.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T21:05:07.000Z","created_at_i":1794258307,"id":50432120,"options":[],"parent_id":50431301,"points":null,"story_id":50429024,"text":"Sorry but why is my reaction "noise"? It's legitimate to be overwhelmed by the amount of released updates and being unable to tell the difference between them.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T20:14:21.000Z","created_at_i":1794255261,"id":50431301,"options":[],"parent_id":50431112,"points":null,"story_id":50429024,"text":"Thanks, in that case why not just not comment instead of adding more noise to the discussion?","title":null,"type":"comment","url":null},{"author":"PaulKeeble","children":[{"author":"dingaling911","children":[],"created_at":"2026-11-09T21:22:46.000Z","created_at_i":1794259366,"id":50432368,"options":[],"parent_id":50431700,"points":null,"story_id":50429024,"text":"If your current model is doing fine, why do you need to look at anything new?","title":null,"type":"comment","url":null},{"author":"dvaplima","children":[],"created_at":"2026-11-09T22:09:02.000Z","created_at_i":1794262142,"id":50432963,"options":[],"parent_id":50431700,"points":null,"story_id":50429024,"text":"indeed, im working with several agents scanning the web and keeping up","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T20:37:02.000Z","created_at_i":1794256622,"id":50431700,"options":[],"parent_id":50431112,"points":null,"story_id":50429024,"text":"Its a full time job to keep up with them all and assess their value in your own work at the moment. Just so many models coming out.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T20:03:05.000Z","created_at_i":1794254585,"id":50431112,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Kind of burnt out on all the model updates at this point.","title":null,"type":"comment","url":null},{"author":"syntaxing","children":[],"created_at":"2026-11-09T20:32:39.000Z","created_at_i":1794256359,"id":50431618,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"What does smallest mean? Even smaller than Muse Glimmer which I doubt? Though it is interesting to see US frontier labs trying to push models that outcompete deepseek flash 1017. It\u2019s a good move from the Deepseek federal government ban. Both Inkling 2 small and this seems to try to fill that gap cost vs performance wise. But I like inkling 2 small more due to open weight even though it has a restrictive license.","title":null,"type":"comment","url":null},{"author":"Retro_Dev","children":[{"author":"loeg","children":[{"author":"kristianp","children":[{"author":"anon373839","children":[],"created_at":"2026-11-09T23:14:12.000Z","created_at_i":1794266052,"id":50433715,"options":[],"parent_id":50433363,"points":null,"story_id":50429024,"text":"It was also nerfed to the point of no longer being in a comparable intelligence class to the original.","title":null,"type":"comment","url":null},{"author":"loeg","children":[],"created_at":"2026-11-10T00:04:41.000Z","created_at_i":1794269081,"id":50434247,"options":[],"parent_id":50433363,"points":null,"story_id":50429024,"text":"We don't know how many parameters the closed Meta models have, so what are you comparing this number against?","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T22:44:38.000Z","created_at_i":1794264278,"id":50433363,"options":[],"parent_id":50432210,"points":null,"story_id":50429024,"text":"Which was 816B parameters.","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T21:11:17.000Z","created_at_i":1794258677,"id":50432210,"options":[],"parent_id":50431857,"points":null,"story_id":50429024,"text":"Spark Open was just October. Give it time. https://research.meta.ai/blog/muse-spark-open","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T20:46:45.000Z","created_at_i":1794257205,"id":50431857,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"When they said the model was the "smallest" of their models yet, I was really hoping it would be open weights. I can't find any weights, how sad :(","title":null,"type":"comment","url":null},{"author":"ProblemExplorer","children":[{"author":"dvaplima","children":[],"created_at":"2026-11-09T22:09:41.000Z","created_at_i":1794262181,"id":50432969,"options":[],"parent_id":50432136,"points":null,"story_id":50429024,"text":"thanks for the compilation of benchmarks","title":null,"type":"comment","url":null}],"created_at":"2026-11-09T21:06:40.000Z","created_at_i":1794258400,"id":50432136,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Meta seems to have designed Ember to be a strong "all round" great value model - it dominates most of its class on cost
The noncoding results on some reports grabbed my attention. Ember also looks like a particularly inexpensive option for some agent workflows.
On Artificial Analysis, Ember max scored 36 on the Intelligence Index at ~$0.14 per task, compared with 35 at $8 for Grok 4.7 xhigh, and 28 at $0.15 for GPT-6 Luna max. Haiku 5.5 max also scores 36, at $0.41 per task. These are AA's measured costs across its index for those specific configurations.
That puts Ember in roughly Grok's aggregate score range at about 1/55 of the measured cost, and ahead of Luna at a similar cost! [1]
The Vals finance/legal results are also particularly interesting: comparing Ember with GPT-6.1 Terra, both evaluated at "max" reasoning:
- Finance Agent v2.1: 63.83% vs 55.96%, at $0.06 vs $1.92 (!) per test. [2]
- Legal Research Bench: 49.04% vs 25.96%, at $0.06 vs $3.22 (!) per test. [3]
- HLAB-Vals-1.1: 25% vs 1.67%, at $0.12 vs $1.48 (!) per test. [4]
Those are INSANE score differences, at roughly 1/32, 1/54 and 1/12 of Terra's respective benchmark costs!
Misleadingly, their overall Vals Index scores are almost identical: 60.48% for Ember and 60.54% for Terra. Terra also scores higher on Vals' Terminal-Bench 4.0 run: 49.49% vs 39.39%. They arrive at similar aggregate scores through quite different strengths. [5][6]
For me, this reinforces the case for custom, workload specific evals. Broad benchmarks have been useful for shortlisting models but choosing between them, calls for representative tasks, explicit success criteria and measurements using the tools and reasoning settings you actually intend to deploy.
I certainly want to measure cost per successfully completed task, including retries and escalation.
Reusing a framework such as Harbor with a custom task suite seems increasingly useful here. [7] The aggregate score here gives me a starting point but the tasks I actually need done then determines my choice.
Goign by the generic benchmarks would confuse me to no end and likely burn me out.
Great times to be in!
[1] https://artificialanalysis.ai/leaderboards/models
[2] https://www.vals.ai/benchmarks/fabv2-1
[3] https://www.vals.ai/benchmarks/legal_research
[4] https://www.vals.ai/benchmarks/hlab
[5] https://www.vals.ai/models/meta_muse-ember
[6] https://www.vals.ai/models/openai_gpt-6.1-terra
[7] https://github.com/harbor-framework/harbor","title":null,"type":"comment","url":null},{"author":"system2","children":[],"created_at":"2026-11-09T21:44:25.000Z","created_at_i":1794260665,"id":50432680,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"When a statement starts with "our," I always expect bullshit. Batch pricing is two times cheaper with Haiku. Benchmarks are nearly tied. Nobody uses the dumbest models for coding either.
If you want customers to switch over, you better make the price at least half of your competitor's.","title":null,"type":"comment","url":null},{"author":"mgdev","children":[],"created_at":"2026-11-09T22:01:42.000Z","created_at_i":1794261702,"id":50432883,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"I can't wait for Muse Smolder. It's gonna somehow run on an ESP32 compatible SOC w/a custom TPU at 100 tok/sec and people are still gonna drag it.","title":null,"type":"comment","url":null},{"author":"ImPostingOnHN","children":[],"created_at":"2026-11-09T22:38:07.000Z","created_at_i":1794263887,"id":50433282,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Where exactly can we download this small model?
How do we know it is small? How many bytes does it consume on-disk and in-memory? Can I run it on an esp32?","title":null,"type":"comment","url":null},{"author":"doodlesdev","children":[],"created_at":"2026-11-10T00:02:25.000Z","created_at_i":1794268945,"id":50434219,"options":[],"parent_id":50429024,"points":null,"story_id":50429024,"text":"Interesting to notice that the best benchmarks of the model are not on Meta's own agent harness, but instead on mini-swe-agent:
> 1. All models run at max reasoning effort. Artificial Analysis results are from Intelligence Index v5, as published November 9, 2026; cost per task uses each model's list pricing. Internal results are from Meta's evaluation suite. The Personal Superintelligence Index (PSI) is an internal composite of eight evaluations: Agents' Last Exam, AutomationBench, FocusBench, JobBench, OfficeQA Pro, GDP.pdf, GDP.xlsx and VBT. Terminal-Bench 4.0 uses each model's best published harness: mini-swe-agent (Muse Ember), Claude Code (Claude Haiku 5.5), and Codex (GPT-6 Luna). OSWorld 2.1 uses partial-credit reward; PerceptionBench is scored with tools.