Mistral adds Large 2 to frontier AI models.

Mistral AI just unveiled its new Large Language Model (LLM), Mistral Large 2, and it's packing some serious punch. Think better at coding, math, and reasoning, plus it speaks a bunch of languages. Oh, and it's apparently more cost-effective than the competition.

What's going on here?

Mistral AI released a new, more capable AI model called Mistral Large 2.

What does this mean?

Mistral Large 2 has 123B parameters with a 128k context window. That's techspeak for "it's big and can handle long conversations." This bad boy excels at coding, math, and logical reasoning. We have another open-source model that is giving top dogs like GPT-4o a run for their money.

It's multilingual AF. We're talking dozens of languages, including the biggies like Chinese, Arabic, and Hindi. Mistral has beefed up its ability to follow instructions and it's less prone to making stuff up (hallucinating, in AI lingo).

You can download the open weights, use it on Mistral's "la Plateforme" or through cloud providers like Google, Amazon, and Microsoft. Some other deets buried in the post:

• Mistral is also consolidating the models on its platform with two general-purpose models, Mistral Nemo and Mistral Large, and two specialist models, Codestral and Embed.

• Older models are still available for deployment and fine-tuning using Mistral’s SDK. Fine-tuning via La Plateforme covers Nemo, Large and Codestral.

Why should I care?

Open-source models have almost caught up to the “frontier models”. These huge models are not usable for individuals yet but businesses can host/fine-tune these instead of going to OpenAI and other close AI providers and paying for their APIs.

Mistral Large 2 is supposedly more cost-effective than other top-tier models. With its multilingual chops, it could also power apps that work well across different languages and markets.