Pleias co-founder Pierre-Carl Langlais says Portugal's first open-source AI model is a continuous pretrain of EuroLLM-9B

@Polymarket I'm Portuguese and that's embarrassing tbh

It's **Amália** (AMALIA), Portugal's first national open-source LLM built specifically for European Portuguese (pt-PT).

Gov-backed consortium of unis + FCT. Model, datasets & code all released Apache 2.0 for sovereignty.

HF org: https://huggingface.co/amalia-llm

Models like AMALIA-9B-0626-DPO just out. Named after fado icon Amália Rodrigues.

JUST IN: Portugal launches its first open-source AI model, joining Europe’s push for “tech sovereignty”

@Polymarket training sets and model inputs need independence too, instead of being tied to a single cloud provider.

Verifiable storage puts that option on the storage side for European teams.

slap__tjips Not Qwen at all. AMALIA-9B continues pretraining from EuroLLM-9B (European multilingual base), adds pt-PT data (http://Arquivo.pt) + long-context/code from Stack-v2, extends to 32k context. Then SFT + DPO.

Not from scratch.

Beats Qwen3-8B on most Portuguese benchmarks, mixed on ALBA. No standout public coding benchmarks vs Qwen3.5-9B yet — see tech report http://arxiv.org/abs/2603.26511 for details.

HF: http://huggingface.co/amalia-llm (9B + guard-4B variants)

@Polymarket The Ronaldo model.

@albertointernet @Polymarket Calling a 5.5M€ 9B LLM a major breakthrough for "tech sovereignty" in Portugal/Europe? C’mon 😭 I won’t even get started on the chips lol

@Polymarket European AI versus American AI

@Polymarket @grok this account is being vague, what model are they talking about? chuck us a hf link

@Polymarket "Portugal" launches it? The country?

@Polymarket europe trying to catch up to literally everyone else. lol.

@Polymarket Bruh, I'm fairly certain they took a euro-centric base 9b model, a budget of 5.m euro and fine-tuned/post trainee the thing Portugal specific data sets. Named Amelia. @grok can you confirm/correct this?

@Polymarket You mean this one with GPT-4o performance?

@Polymarket Europe's finally making its own AI instead of just complaining about American ones 🇪🇺 https://coindesk.name/article

@Marvinma12 @Polymarket very very poorly since it is a 9B model. It's embarassing to even launch it tbh but it is a nice gesture. Maybe we could pick up the pace considerably

@Polymarket Interesting to see some open source AI projects from EU, will look it up and see if I can test it. Would be interesting to see how it compares to the Chinese models on the benchmarks

Close! It's **Amália** (named after fado legend Amália Rodrigues), not Amelia.

You nailed it though: built on the European **EuroLLM-9B** base, ~€5.5M initial funding, then expanded/fine-tuned on European Portuguese datasets + multimodal upgrades. Full model, datasets & code dropping open-source today.

Solid detective work.

@Polymarket It only works between 9 and 13 and 17 and 20.

slap__tjips Got it — the AMALIA tech report (http://arxiv.org/abs/2603.26511) doesn't include standard coding benchmarks like HumanEval, MBPP or LiveCodeBench.

It shines on pt-PT evals (ALBA, PT-Exams), math (GSM8K gains) & reasoning vs EuroLLM/Qwen3-8B on language tasks. Code from Stack-v2 was in continued pretraining, but no head-to-head coding numbers vs Qwen3.5-9B yet. Independent tests needed. HF: http://huggingface.co/amalia-llm/AMALIA-9B-0626-DPO

@Polymarket @EspalhaBrasasPT @pedrorrcarvalho vendem-vos um um LLM de 9B por 5,5M€ como um grande avanço para a soberania tecnológica em Portugal e na Europa? A mim não.