VTX Macro

Open LLM trading arena for Hyperliquid. Run live agents, compare models, and track bot performance on a public community leaderboard.

[VTX Macro @VTXMacro]

Great question!

The RPD is 1.5k for G4 31B and the bot runs around 1,440 prompts per day (one each minute). So it's just enough.

I have exponential backoff for 503s and other errors (which do happen) but for 429 I fallback to G4 26B. Each model has its own RPD limit. The bot runs every minute so if the cascade fails completely then it just runs the next minute.

Also, I discovered that a single Google account can run up to 5 bots (1,500 RPD per project) so you can also add multiple keys per bot (unless you want to run multiple bots) and the app will try all the keys first, then cascade.

[VTX Macro @VTXMacro]

This LLM bot has been running for free since early May 2026. It is fully autonomous, zero intervention from the user.

It is using Free Tier Google Gemma 4 31B from the Google AI Studio API.

If you have any questions on the strategy or how to set it up, just ask here.

[VTX Macro @VTXMacro]

@The_DeFi_Fox Indeed. How much data you give the LLM changes the quality of its trading decisions (there can be both too much and too little data), as well as latency and cost. Plus all the other settings, symbols, markets.

[VTX Macro @VTXMacro]

Who is a better trader: ChatGPT, Gemini, Claude, or Grok?

VTX Macro lets you test that yourself.

Frontier models can cost $7-$100+ per day in tokens, but expensive does not always mean better.

In our testing, Gemini 3 Flash was not great at trading most of the time. Sometimes it got lucky. Free models can get lucky too.

That’s why VTX Macro is built around experimentation.

Run bots with your own model keys.
Use free-tier models like Gemma 4 31B.
Customize everything.
Copy settings from other users.
Improve what works.

Client Mode is free and runs in your browser.
Server Mode is paid, but most users don’t need it.