frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model

https://github.com/nex-agi/Nex-N2/issues/4
56•unrvl22•1h ago

Comments

unrvl22•1h ago
The municipality of Rio de Janeiro (via its IT company IplanRIO) released Rio-3.5-Open-397B, presented as a homegrown Qwen3.5 fine-tune that beats comparable open models on benchmarks. The linked issue argues it's actually a weighted merge of ~60% Nex-N2 Pro + ~40% Qwen3.5-397B-A17B - Nex-N2 having been released about a week earlier.
elzbardico•49m ago
This is so typical of brazilian academia.
guiraldelli•40m ago
Without evidence, your comment is just bad mouthing.

I have been involved in academia, including in Brazil, and I don't find academia there any more copycat than any other institution, including top tier ones.

_3u10•38m ago
No, typically Brazilians go to Paraguay for their education, most of their technology comes from Paraguay too.
cassiogo•22m ago
What? Never heard of this
dghlsakjg•8m ago
This was a municipality working with a government associated IT company.

What does it have to do with Brazilian academia?

AnotherGoodName•45m ago
This is fascinating that it worked though. Can we just merge all the open weight models and get something better?
_3u10•39m ago
No, they need the same arch, but you can distill them into a single model. And yes, if you use the API directly Claude will often say it’s an open weight model (likely the ones it was distilled from)
wds•34m ago
I imagine it'd work the same as merging all the good-tasting foods to get an even tastier one
dindunuf•18m ago
that kinda worked in llama 1/2 era, not between different models but between finetunes of the same model. the briefly legendary Mythomax was IIRC a merge of 5+ tunes, some of which were merges themselves.
avereveard•12m ago
most merge improve a small subset of "feeling" benchmark (too small, too specific, or out of distribution) and tend to show degradation on actual benchmark, with especially punishing result on long chain benchmarks.

also only work on matching architectures (i.e. finetunes/loras of the same model)

AlienRobot•38m ago
The model's webpage at https://huggingface.co/prefeitura-rio/Rio-3.5-Open-397B says it's a merge now. It previously didn't contain this paragraph:

>The model is built via a merge of https://huggingface.co/nex-agi/Nex-N2-Pro and https://huggingface.co/Qwen/Qwen3.5-397B-A17B, proceeded by On-Policy Distillation from a stronger model. We detected an incorrect upload in the previous version, where the base merged version was upload instead of the final distilled model. We are sorry for the confusion and apologize profusely.

Incidentally are people using Github issues as blogs now?

zinodaur•37m ago
Oh no, someone is profiting off of their work without proper attribution!?!?
internet2000•35m ago
Attribution isn't the relevant part. Lying about your lab's capabilities is.
Planktonne•27m ago
That's also something all the AI companies have been doing.
dofm•10m ago
Lying about model capability is right now the lingua franca of the cloud AI business model, almost; they yes-and each other's lies because they are in a position of needing to generate interest, including going as far as needing to trigger regulatory capture.

(It's not news to anyone who has worked in sales-led businesses that salespeople are prone to believing the claims of other salespeople, I guess).

functionmouse•16m ago
leopards ate my face
adrian_b•16m ago
I do not see anyone lying.

The model card says:

> Post-trained from Qwen 3.5 397B

The model card also says that they use an inference framework based on "SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMs" by Shi et al.:

https://arxiv.org/abs/2510.05069

So the sources seem properly attributed.

They only claim that what they did to "Qwen 3.5 397B" has improved the LLM, including, as expected, with "strong performance in Portuguese".

alfiedotwtf•32m ago
Wasn’t it already obvious given the awfully familiar parameter numbers?
ekjhgkejhgk•25m ago
One funny thing about incompetence is that they don't have the competence to know that their incompetence is straightforward to verify by a competent person.
root-parent•20m ago
You just described every single vibe coder...
carlosjobim•8m ago
Why would they care? They get their salaries and pensions and bonuses, and the tax payer is footing the bill.
MadrasTh0rn•25m ago
Not surprised
fkozlowski•21m ago
I'm honestly surprised that they even had the inclination to attempt creating a model. I guess it's bullish that a municipal IT department had the guts to try this?
yieldcrv•18m ago
Didn’t the last thread about this have someone from the lab or an enthusiast in Rio saying exactly that?

Its a fine tune of Qwen

Not a conspiracy

bachmeier•16m ago
"Their work"? First you had the original content creators that did 99.99% of the work. Then you had the US companies bundle it up into a frontier LLM. Then "they" did the "work" of using the US model as a foundation for their own. So in the sense of doing 0.00001% of the actual work that went into their product, sure.

I'd say it's more like someone forking a Linux distro, adding a few themes and fonts, and then complaining when someone else forks their distro and adds another theme.

dghlsakjg•10m ago
That’s the joke.
harikb•9m ago
It is only a problem if you claim it to be an independently developed OS with no attribution to base
bwilliams18•8m ago
That was the joke of the parent comment.
JoshStrobl•6m ago
That joke really went over your head, huh...
woadwarrior01•12m ago
Are you new to the latest AI hype cycle? /s
carlosjobim•11m ago
This is a pure scam on tax payer money. But what else would be expected?

Soviet Whaling

https://slate.com/news-and-politics/2022/06/history-of-soviet-whaling-greenpeace-twentieth-centur...
1•simonebrunozzi•52s ago•0 comments

Multistack – Lightweight TUI for orchestrating coding agents

https://github.com/gi-dellav/multistack
2•gidellav•4m ago•0 comments

Zerostack v1.5 – Lightweight Unix-inspired coding agent

https://github.com/gi-dellav/zerostack/releases/tag/v1.5.0
2•gidellav•4m ago•0 comments

The Road to a Social Cybernetics

https://miltonlmueller.substack.com/p/the-road-to-a-social-cybernetics
2•dimiprasakis•6m ago•0 comments

Cull–screenshot widget for Windows (Ctrl+PrtScn → save/copy/delete in 7 seconds)

https://www.cull.live/
3•dev1601•8m ago•0 comments

Knowing How and Knowing That – Gilbert Ryle [pdf]

https://www.informationphilosopher.com/solutions/philosophers/ryle/Ryle_KnowHow.pdf
2•soupspaces•10m ago•0 comments

Is it possible to write a kernel module which will blow the PC speaker?

https://lore.kernel.org/lkml/CABG1boPZkP_HxOY+96cKqrv9UKujWnaWK-_dPQX+Zk6BSuZrwA@mail.gmail.com/
2•thinkingemote•11m ago•0 comments

JPMS Explained Through a C# Analogy

https://old.reddit.com/r/java/comments/1u5o74m/jpms_explained_through_a_c_analogy/
2•Tomte•12m ago•0 comments

Topological Derivation of Toronto's Time Boundaries (4/3πC)

https://medium.com/@f9121212/topological-derivation-of-geometric-boundaries-for-positive-and-nega...
2•ortrich•13m ago•0 comments

Plugins Case Study: Pluggy

https://eli.thegreenplace.net/2026/plugins-case-study-pluggy/
2•ingve•13m ago•0 comments

Automation modes in Home Assistant: Why the default isn't your friend

https://frenck.dev/automation-modes-in-home-assistant-why-the-default-isnt-your-friend/
2•RyeCombinator•13m ago•0 comments

Dillo directory – Directory of useful sites that work reasonably well on Dillo

https://dir.dillo-browser.org/
4•HotGarbage•15m ago•0 comments

The Strait of Hormuz Has Been Closed for 100 Days. Why Aren't Oil Prices Higher?

https://www.wired.com/story/strait-of-hormuz-closed-100-days-why-arent-oil-prices-higher/
6•joozio•16m ago•1 comments

IdeasBerg has analyzed 30 SaaS ideas from Greg Isenberg's content

https://ideasberg.com/business-ideas/saas
2•pro_methe5•25m ago•0 comments

Apple's Private Cloud Compute Is Severely Limited for Third-Party Developers

https://developer.apple.com/private-cloud-compute/
5•Brajeshwar•27m ago•2 comments

I automated the part of freelancing nobody talks about

https://www.leadlu.com/
2•brevn•29m ago•0 comments

Starmer to announce 'Australia plus' ban on social media for under-16s

https://www.theguardian.com/uk-news/2026/jun/14/starmer-to-announce-australia-plus-ban-on-social-...
4•c-oreills•29m ago•0 comments

The Story of PHP. Documentary Teaser [video]

https://www.youtube.com/watch?v=4W4y46WVdCU
3•pjmlp•30m ago•0 comments

Periphery Alignment and The 2 body hypothesis

https://zenodo.org/records/20691150
2•KridayDave•30m ago•0 comments

Fifteen short films about AI-era threats, made with Conan

https://www.adaptivesecurity.com/conan
1•mooreds•30m ago•0 comments

Bod 26-04: Prioritizing Security Updates Based on Risk

https://www.cisa.gov/news-events/directives/bod-26-04-prioritizing-security-updates-based-risk
1•mooreds•32m ago•0 comments

France probes compromise of gov messaging platform after account hijack

https://www.theregister.com/security/2026/06/09/france-probes-compromise-of-gov-messaging-platfor...
3•mooreds•32m ago•0 comments

David Sacks on Anthropic export control

https://twitter.com/DavidSacks/status/2065853007619588171
4•satvikpendem•36m ago•2 comments

Local .NET multi-agent LLM pipeline AI therapist using a compact wire format

https://github.com/paulomac1000/hand-codechttps://github.com/paulomac1000/hybrid-therapist-ai
1•paulomac1000•36m ago•0 comments

Structured Subagents for Claude Code

https://github.com/wastedcode/truecast
1•zeppelin_7•39m ago•0 comments

Building a Tiny FUSE Filesystem

https://www.shayon.dev/post/2026/161/building-a-tiny-fuse-filesystem/
1•shayonj•40m ago•0 comments

Quieren meter pagos dentro de ChatGPT

https://substack.com/profile/467902996-juanlu-de-apisdom/note/c-276207917
1•ApisDom•42m ago•0 comments

Show HN: Peek – A Figma like DB GUI

https://getpeek.dev
2•tehrash•42m ago•0 comments

Just-World Fallacy

https://en.wikipedia.org/wiki/Just-world_fallacy
3•moritzwarhier•44m ago•1 comments

Etechx Centre

https://etechx.co.ke/i-waited-for-a-man-who-was-never-mine
1•ndegekm•46m ago•0 comments