Fractal 2.0

This is the last blog post before the upcoming Fractal 2.0 announcement. It outlines all the numerous tech updates of the AIKEK platform. It finalizes Fractal 1.0 development and lays the ground for v2.0.

Decentralized Reinforcement Learning (DeRL)

Decentralized AI training will play a huge role in Fractal 2.0. In the meantime, DeRL Widgets - the first proof-of-concept of holders-aligned AI training technology - began to roll out.

DeRL Widgets

The newest version of the Prompt Enhancer is fine-tuned by DeRL widgets - a new type of extremely smart AI that learns from AIKEK holders in real time.

Under the hood, it works as a scaffold for the online RL, which is designed to train the ASI that is aligned with AIKEK holders.

It leverages the new tokenomics: votes by bigger $AIKEK holders are more valuable and create a bigger impact on AI training.

More RL widgets are in development: for data, for memes, and for completely novel use cases.

AIKEK + Robotics = Openfren

Openfren is your proof-of-concept open-source robot fren for Internet Capital Markets.

It uses the Alphakek Fractal knowledge engine as its brain.

The goal of Openfren is to establish the baseline platform for AIKEK ⟷ robotics interaction prior to Fractal 2.0 release to make it easier for Fractal 2.0 to onboard robotics developers.

Compute Credits

The new compute credits system streamlines the utility of the $AIKEK token:

• Earn credits by holding AIKEK
• Spend them by using AIKEK tools
• Share credits with other users and groups

x402 Support

For B2B users and developers who want to get a higher Fractal API quota than possible by just holding tokens, an x402 API gateway is now available.

Video V2

Video v2 is an entertaining way to battle-test two new features:

1. GPU scheduler for compute-intensive tasks;
2. Turning the entirety of AIKEK into one online RL environment - AI that learns from you in real time.

Infrastructure Updates

During 2025, AIKEK experimented with various thinking and non-thinking models to find the balance between speed and accuracy. During the previous update, AIKEK's underlying LLMs were switched to non-thinking mode to make it faster and more responsive for the end user (APIs, TG Playground, etc).

The new update makes AIKEK's AI inference much faster, and in turn lets the models reason more while taking as much time to respond. The result is: models are bigger, smarter, and just as fast as before.

AIKEK will turn AI training into gambling

Fixes & Small Features

Multimodal Mode

Multimodal mode now supports multiple images at once → combining and mixing memes and data sources became possible:

Context Engineering

Context handling in the Telegram playground has become much smarter, including chat history, images, videos, GIFs, links, etc. It lets the AIKEK Agent be much more responsive and useful.

Fixes

• Fixed a bug where voting on an image resulted in a false "You already voted!" message
• Fix compute credits sometimes being incremented too slowly
• Fix AIKEK Playground occasionally showing an error when analyzing a PDF file
• Fix rare error when analyzing a token that has multiple LPs

About Alphakek AI

AIKEK builds AI learning environments for compounding capital onchain. Power billions of agents, robots, and prediction markets. Featuring unbiased, crypto-aware AI and an NVIDIA-recognized knowledge layer powering crypto's biggest names.

Twitter | Telegram | Website | Discord | CoinGecko | CoinMarketCap