SOTA Open Source Kimi K2 released!

Dear Readers,

Sometimes events unfold at breakneck speed – and that's exactly how the current AI world feels. Kimi K2, a Chinese model, is stepping into the spotlight and openly challenging what we previously considered to be the pinnacle of AI. With 1 trillion parameters, a novel optimizer, and capabilities that make GPT-4.1 sweat, open models are dangerously approaching what only closed-source giants have mastered so far. But while Moonshot AI is electrifying the scene, Sam Altman is putting on the brakes – a coincidence? Or the calm before the storm?

In this issue, we dive deep into the shifts: from Kimi K2's impressive benchmarks to the question of whether AI is already deciphering the ECB's monetary policy, to the credit market revolution being driven by fintechs such as Abound. And we ask the provocative question:

Can AI really pick your stocks – or is it just a smarter assistant?

If you want to know why these developments are more than just headlines, read on – the answers will surprise you.

In Today’s Issue:

All the best,

Moonshot AI has released Kimi K2, a new large model that is electrifying the AI community. Using a mixture-of-experts approach (32 billion active parameters out of a total of 1 trillion), Kimi K2 combines extreme computing power with remarkable efficiency. It has been trained on over 15.5 trillion tokens and uses the innovative MuonClip optimizer, which enables stability and scaling. The community now has access to two variants: Kimi-K2-Base for its own experiments and Kimi-K2-Instruct for ready-to-use chat and agent functions.

The benchmark results are particularly impressive: On SWE-bench Verified, Kimi K2 achieves 65.8%, surpassing all previous open-source models and coming dangerously close to Claude 4 Sonnet. On LiveCodeBench, it delivers 53.7% – significantly above GPT-4.1 and DeepSeek-V3. It also performs strongly in mathematical disciplines such as AIME and MATH-500. Unlike classic language models, Kimi K2 is also optimized for autonomous use as an “agent”: It can control tools, write code, execute commands, and solve complex tasks independently.

At the same time, Sam Altman has announced that he will be postponing his own open source model, citing security concerns. A connection to Kimi remains conceivable, however.

Why it matters: Kimi K2 marks a turning point in open source, putting GPT-4-level capabilities in everyone's hands and opening up new possibilities for developers to build powerful and flexible AI agents.

Sources:

🔗 https://www.marktechpost.com/2025/07/11/moonshot-ai-releases-kimi-k2-a-trillion-parameter-moe-model-focused-on-long-context-code-reasoning-and-agentic-behavior/

🔗 https://huggingface.co/moonshotai/Kimi-K2-Base

This is the easiest way for a busy person wanting to learn AI in as little time as possible:

Sign up to start learning.

An AI analysis by DIW Berlin increases the accuracy of ECB policy forecasts from ~70% to ~80%. It systematically evaluates language signals from ECB communications, combined with macro data, and sees a high probability of an interest rate cut.

Economic relevance: This turning point shows how AI increases transparency in the monetary policy signaling process. Central banks could change their communication strategies, markets would react faster and in a more informed manner – a structural change in monetary policy transmission.

The London-based fintech Abound increased its profits from £0.3M to £7.5M thanks to Render – an AI-supported lending model. It has seen £900M in loan volume since 2020 and 151% revenue growth. Render analyzes bank transactions, assesses risks with greater precision, and even supplies its technology to banks themselves.

Economic relevance: AI is becoming a central technology in lending, shifting risk and business models. Banks will license artificial intelligence instead of developing it themselves – a signal of disruption in the European credit market.

A quantitative professional warns that LLMs such as ChatGPT lack real-world experience and intuition – AI is currently only suitable as an assistant for analysis and text editing. It will not replace investors, but tools for data interaction such as prompt terminals are gaining in importance.

Economic relevance: AI is not triggering a paradigm shift in investment management – but it is changing the infrastructure. Industry players are investing in AI terminals instead of changing strategies and restructuring analysis processes.

Working on the future of finance through AI? From trading algorithms to credit modeling and fintech infrastructure, we’re looking for work that explores the intersection of AI and financial systems.

Submit your paper or project to Superintelligence, the top AI newsletter with 200k+ readers, by emailing [email protected] with the subject line “Finance Submission”. We’ll contact you if we’d like to feature it.