A brief discussion on Launchpad for AI large models

Author: Haotian; Source: X, @tmel0211

After listening to the FLock 2025 annual performance report, I was particularly intrigued by a mention of building a Launchpad for AI large models.

What? Another Launchpad? How do large models issue assets? Actually, it’s quite understandable—think of it as an analogy:

A Launchpad for AI Agents like Virtuals Protocol, driven by the application layer, uses token incentives to send assets to Agents, helping them evolve from “chatting” to x402 “paying,” and ultimately to the goal of “autonomous trading” and providing complex services;

Meanwhile, FLock’s plan to create an AI Model Launchpad is infrastructure-driven, issuing assets to trained large models, specifically many vertical domain models, such as medical diagnostics, legal documents, financial risk control, supply chain optimization, and more.

While these vertical models have relatively controllable training costs, their commercialization paths are very narrow—they either sell out to big corporations or go open-source for passion projects, with few sustainable monetization methods.

FLock aims to use Tokenomics to reconstruct this value chain, issuing assets to fine-tuned large models, thereby enabling data contributors, compute nodes, validators, and others involved in model training to potentially earn long-term benefits. When a model generates income through calls, rewards can be continuously distributed based on contribution ratios.

The idea of building a Launchpad for large models sounds fresh at first glance, but essentially, it’s about using financialized methods to drive product development.

Once the model is assetized, trainers will have ongoing motivation to optimize, and if earnings can be continuously distributed, the ecosystem will develop self-sustainability.

The benefits of this approach are undeniable. For example, the recently popular NOF1 large model trading competition currently only features general large models; specialized fine-tuned models are not competing because there’s a lack of incentive mechanisms. Excellent specialized models tend to stay hidden for profit, making exposure unlikely. But if assets are involved, it’s a game-changer. This kind of large model Arena becomes a public stage to showcase capabilities, and the competitive performance directly impacts the asset value of the models, opening up a realm of possibilities.

Of course, FLock has only proposed a direction so far and has not yet been fully implemented. The specifics of issuing assets to models versus issuing assets to Agents remain to be seen.

But one thing is certain: how to ensure that model calls for asset-issued models are based on genuine needs rather than spam, and how to effectively ensure product-market fit (PMF) within vertical scenarios, are crucial issues. It’s safe to say that the problems faced by the Agent token issuance wave will also be relevant here.

I’m just really looking forward to seeing what different gameplay approaches might emerge from building a Launchpad for Models.