tl;dr

• How wild is Grok 4.3?
• Mistral’s struggles continue

Releases

Grok 4.3

xAI (now a part of SpaceX) has quietly released Grok 4.3

The model falls into the mushy middle with the likes of Meta’s Muse Spark and the plethora of Chinese models. It is half a trillion parameters, and is around Sonnet 4.6 level on benchmarks, but based on the previous Grok models it is probably a bit overfit and will be behind Sonnet in real world use.

Grok 4.3 struggles relative to its Chinese rivals when it comes to coding and agentic tasks. It does shockingly well on the Vals AI, being the best model on their corporate finance and case law benchmarks, beating every other model (GPT 5.5 and Opus 4.7 included).

If you are using AI for finance applications or case law, then I would give this model a look, otherwise it is a pass for now.

It will be interesting to see how xAI continues going forward, as they have lost every co founder (except Elon Musk) along with a large number of their top engineers.

Quick Hits

Mistral Medium 3.5

Mistral is still struggling to release a high quality model, with their 128 billion parameter dense “flagship” Mistral Medium 3.5 model being big, slow, expensive, and no better than similarly sized MoE models.

Finish

I hope you enjoyed the news this week. If you want to get the news every week, be sure to join our mailing list below.