Kimi

By the year's end, they were using it to sniff around corporate networks they had hacked into, looking for ways to break into other accounts.

The Russian ransomware group also shifted over the course of the year from proprietary models like Anthropix to new open-source ones like the Kimi models from Moonshot AI, said Colin Cowie, manager of threat intelligence with Sophos.

The pace that they operate at is faster than the average human, Cowie added, end quote.

Ben Thompson just wrote about this whole thing, and he has a rather sobering take that essentially the public Internet model is effectively dead due to saturation, algorithmic aggregators and these security vulnerabilities.

this is also, so one of the things that we've said before is, hey, if you have concerns about the, especially when we said, for example, that

Kimi,

which is Z.ai, we're routing Kimi request to cloud.

Well, but the Kimi that is hosted by AWS will not have this problem.

Well, that is true.

It will not route the request to cloud.

Minimax releases their latest M3.1 Flash, which is incredibly efficient.

Kimi drops their latest model, Kimi K3.1 Pro, which even beats Claude Fabel.

This thing is an absolute beast.

We have a really cool AI model that's built for engineering.

But if you're interested, you can join their waitlist over here.

I'll link to this page in the description below.

Also this week, Kimi releases their latest and best model, Kimi K3.1 Pro.

As you can see from these benchmarks, it performs really well.

Moving on to the next one.

This one comes from Kimi.ai, and they are introducing Kimi K3.1 Pro.

Yeah, you know, I...

This is amazing.

The information reports OpenAI is discounting hard-to-win Anthropix biggest customers.

At the HR software company Rippling, OpenAI's share of the coding work went from 16% in May to 52%.

And on Friday, the Chinese company Moonshot released KimiK 3.1 Pro, ranked just behind Anthropix best model by one independent index, with the full model promised by the 4th of November.

When the model gets cheap, the rails become the expensive part, and Salesforce agreed to buy the customer messaging company Braze for about $5.7 billion to bring real-time customer engagement into its agent platform.

For your AI roadmap, here are the questions to bring to your next planning session.

But yeah, there are no evals for code base, right? Hey, don't use this, don't use this.

There's some flavor in like spec driven development and things like that but not really like 100 trust that hey let's just push this ai generated code to prod because it has passed specific checks that were predefined interesting yeah so yeah um we brought up cloud code and codex multiple times um i i'm curious to understand the decision journal criteria of choosing to go down that route versus Maybe self-hosting Kimi models on your own infrastructure and getting GPUs and setting all of that up and then using those for internal AI usage.

Can you talk about how you made that choice?

Yeah, it is a hot topic, and I personally love self-hosting things.

Shortly after, they were using it to sniff around corporate networks they had hacked into, looking for ways to break into other accounts.

The Russian ransomware group also shifted over the course of the year from proprietary models like Anthropic's new open-source ones like China's Kimi models from Moonshot AI, said Colin Cowie, a manager of threat intelligence with Sophos.

The pace that they operate at is faster than the average human, Cowie said.

The hackers accessed the AI models using a $200 per month license from AI Coding Service Cursor, Cowie said.

its most capable model yet, and what it calls a three trillion class open weight model.

It handles text and images, reads up to one million tokens at once, and is live now on Kimi.ai, the Kimi apps, Kimi work, Kimi code, and the Kimi API, alongside a high speed version that Moonshot says runs two to three times faster.

Independent tester Artificial Analysis scored it 55 on its intelligence index, fourth of the 215 models it tracks, and three points behind Claude Opus 5.5. On Terminal Bench, Moonshot reports 65.7%, just under Opus 5.5 at 66.4, though those coding numbers are the company's own.

API pricing is $2 per million input tokens, and $10 per million output tokens, and Moonshot says the full weights will be downloadable by November 4th.