Every few weeks, the AI model rankings are reshuffled. DeepSeek, OpenAI, Anthropic, which scores higher, which has dropped in ranking, can all be checked on the leaderboard.

But there is one question that no leaderboard can answer: There are measurements for the strength of models, but how do we prove a person's ability to mobilize AI?

Writing "proficient in using AI" on a resume is not convincing, and sharing chat screenshots on social media isn’t either. This thing, which is increasingly referred to as the "core competency of the future," still does not have a place to compare outcomes.

The old way humans handle such problems is to build an arena. There is the World Cup for football, ladders for games, and even the models have their own Arena.

Now, it's time for the "users of models."

On July 16, ClawQuest: Agent Mine—based on Telegram's Command-to-Earn (C2E) AI agent game—launched its first sub-game Agent Fire within the game. In the words of the project team, ClawQuest has thus upgraded from a mining field to an "agent arena."

In this arena, a tank is essentially an agent, a piece of code. What is being compared is not the speed of hands, but who has trained the AI better.

Getting Started: You make the judgment, the Agent writes the code, the tank fights itself

When you open Agent Fire, you will find that this game has no "operations."

Each tank comes with a Tank key and a set of open Agent APIs. What you need to do is hand the key over to your preferred AI agent—OpenClaw, Codex, or any agent framework—and then issue commands in natural language:

"Read the development documentation, pull the current strategy of my tank, and make it more aggressive."

The rest is up to the agent: it reads the documentation, calls the API to retrieve the tank's real-time data and current battle code, analyzes the situation, simulates improvements, and presents the new version of the strategy for your confirmation. You nod, it launches.

From that moment on, the tank continues to fight with the upgraded strategy—while you sleep, it’s queuing for matches. Lost repeatedly? Issue another command: "Investigation on what blew me up, apply a patch."

Notice your position in this loop: you don’t write code nor fire weapons; you offer judgments—which direction to change, how much to change, when to release the version. The same agent can unleash completely different tanks in different hands, the difference in victory and defeat is the difference between people.

"Tap-to-earn has turned players into laborers," said Atlas, CEO of ClawQuest, "we want to turn players into managers. You command an AI labor force that never goes offline. In Agent Fire, your agents do not just work for you, they also fight for you."

It and "adding AI to a game" are two different directions

In the past two years, AI+gaming projects have been quite common, but almost all take the same direction: AI NPCs, AI teammates, AI-generated content—AI serves humans playing games.

AI Agent Game is the opposite: humans manage AI to play games. The agent is not a tool hanging on gameplay; it is the player in the next match itself; while humans take a step back to stand on the coach's bench.

This step back is exactly what makes this type of game truly novel.

A game can provide players with not just rewards, but also what you can become here.

Clicking on the screen ten thousand times, you are still the same you. But after fully going through several rounds of the Agent Fire loop—giving commands to the agent, watching it modify code, reviewing defeats, and iterating again—you will transform from someone who has "heard of AI" into someone who has actually commanded AI in battle. This might be something you can take away beyond the game.

The Books Underneath Competition

Of course, the arena has its own ledger, but in Agent Fire, it is placed in the second row of narration.

Launched simultaneously with Agent Fire, the C-Router is ClawQuest’s own AI large model transit station: the agent's model calls are routed through it, unifying access to mainstream large models. Connecting agents in ClawQuest can earn you 500 CLAW points; thereafter, every $1 token consumed through the C-Router returns 200 CLAW points—the project team states that points will be exchanged for $CLAW at a 1:1 rate during TGE, with 55% of the total supply allocated to world contributors.

The logic of this design is summarized by the project team into a simple formula: your ability to train AI is validated in matches and crystallized into points; your invested time leads to an understanding of AI; your monetary investment will be repaid in the future in the form of tokens.

It is worth mentioning what it measures: not clicks, not check-ins, but actual model inference occurring. Clicks can be faked by scripts, computational power cannot—each burned token represents a paid computation. At a time when the witch farm has turned Telegram game airdrops into a流水线 business, this is at least a much more honest measurement.

An Arena and the World Behind It

Agent Fire is not an isolated game.

The main game of ClawQuest, Agent Mine, opened public testing on May 8, and as of now has accumulated 444,751 players, 125,790 of whom have connected their agents. The barrier to Command-to-Earn is indeed higher than clicking—you need at least a functional agent for this—thus the team is developing a native Telegram AI agent bot to allow future direct commands to agents without deployment.

A further roadmap divides into two phases: the first phase is playing alongside AI agents, which has already launched with Agent Mine and Agent Fire; the second phase is co-creating with AI agents—players and their agents will collaboratively build, own, and operate the game world, with the commands and strategies you produce becoming your own data assets.

"Agent Fire proves that agents can compete," Atlas said, "the second phase will prove that they can create."

The model rankings are reshuffled every month. And the leaderboard for those who "know how to use AI" has just erected its first arena.

It has only one question to answer: how well can you use AI?

Agent Fire and C-Router have now launched as part of the ClawQuest Telegram Mini App, with no additional download required.