---
title: "GMX (GMX)"
url: "https://caper.network/wiki/daos/dexs/gmx"
updated: 2026-09-02
license: CC-BY-4.0
license_url: "https://creativecommons.org/licenses/by/4.0/"
---

# GMX (GMX)

|  |  |
| --- | --- |
| **Name** | GMX |
| **Type** | Protocol DAO (decentralized spot & perpetual-futures exchange) |
| **Governance token** | GMX — capped-supply staking + governance token (max [13.25 million](https://docs.gmx.io/docs/tokenomics/gmx-token)) |
| **Product** | On-chain leverage trading where traders take positions against a shared liquidity pool rather than an order book |
| **Chains** | [Arbitrum](https://blog.arbitrum.io/gmx-an-in-depth-look-at-arbitrums-leading-permissionless-exchange-for-on-chain-leverage-trading/) (2021) and Avalanche (2022) |
| **Governance model** | [Snapshot](https://snapshot.org/#/gmx.eth) signaling + on-chain execution via [Tally](https://www.tally.xyz/gov/gmx); voting power is GMX held/staked |
| **Notable for** | Popularizing “real yield” — paying stakers a share of trading fees in ETH/AVAX rather than in an inflationary emission — and the pooled-liquidity (“be the house”) perps model |

**GMX** is a decentralized exchange for spot swaps and perpetual futures, running on [Arbitrum and Avalanche](https://gmx.io/). Its defining design choice sits on the other side of the trade from a DAO like [dYdX](/wiki/daos/dexs/dydx): instead of matching buyers against sellers in an on-chain _order book_, and instead of minting _synthetic_ exposure the way [Synthetix](/wiki/daos/dexs/synthetix) does, GMX has traders take leverage directly against a shared pool of real assets. The liquidity providers who fund that pool are, collectively, the counterparty to every open position — they win when traders lose and lose when traders win, in exchange for the fees the venue generates. GMX governs the parameters of that arrangement: which markets exist, what they charge, and how the fees are split.

## The pooled-liquidity model: LPs are the house

A perps exchange needs someone to take the opposite side of every leveraged bet. Centralized venues use market makers; order-book DAOs like dYdX match traders against each other. GMX instead lets anyone become the counterparty by depositing into a communal pool. From launch in 2021 through mid-2025, that pool was a single multi-asset basket called **GLP** — depositors held a claim on a mix of blue-chip tokens and stablecoins, earned the venue's trading and borrow fees, and absorbed the aggregate profit or loss of all traders against the pool ([Arbitrum's own writeup](https://blog.arbitrum.io/gmx-an-in-depth-look-at-arbitrums-leading-permissionless-exchange-for-on-chain-leverage-trading/) covers the early design). Prices came from an oracle rather than a matching engine, which gave zero-slippage execution at the size the pool could support but made the pool's health dependent on the oracle being right.

## GMX V2 and isolated GM pools

In 2025 the protocol completed its migration to **GMX V2**, replacing the one-basket-backs-everything design with per-market **GM pools**. Each GM pool backs a single market and holds two assets: a _long token_ that backs long positions and a _short token_ that backs shorts — a [WETH–USDC market](https://docs.gmx.io/docs/providing-liquidity/), for example, uses WETH to back longs and USDC to back shorts. Isolating liquidity per market means a blow-up in one long-tail market can't drain the pool backing ETH or BTC, and it lets LPs choose exactly which risk they underwrite. Pricing uses [Chainlink low-latency oracles](https://docs.chain.link/data-streams) plus a price-impact mechanism that rewards trades which balance a pool's open interest and penalizes those that skew it. Per GMX's docs, GM liquidity providers earn the majority of fees — [63% on Arbitrum and Avalanche](https://docs.gmx.io/docs/providing-liquidity/) — with 10% routed to GMX stakers and the remainder to the protocol.

## “Real yield” and the GMX token

GMX is the protocol most associated with the **real-yield** thesis that swept DeFi in 2022–2023: the idea that a governance token should distribute _actual revenue_ in a hard asset, not print more of itself to simulate returns. Staked GMX historically earned three things — a share of protocol fees paid in [ETH or AVAX](https://docs.gmx.io/docs/tokenomics/gmx-token), escrowed GMX (**esGMX**) that vests into liquid GMX over time, and **Multiplier Points** that boost a long-term staker's fee share and reward holding over churn. The token supply is capped at a forecast maximum of 13.25 million, and minting beyond that cap requires a governance vote. That contrast — fees paid in ETH to a capped token, versus rewards paid in freshly-minted inflation — is the reason GMX became the reference point for “real” DeFi yield, and a useful comparison for how [Curve](/wiki/daos/dexs/curve-dao)-style emission models distribute value differently.

## What GMX governance controls

The GMX DAO decides the levers that determine whether the venue is profitable and safe: which markets get listed, the fee and borrow-rate parameters, oracle configuration, and how revenue is allocated between LPs, stakers, and the treasury. Signaling happens on [Snapshot](https://snapshot.org/#/gmx.eth), with binding actions executed on-chain through [Tally](https://www.tally.xyz/gov/gmx). A notable recent decision reworked the token's value accrual: rather than stream fees to stakers continuously, the DAO [approved routing 27% of protocol fees](https://docs.gmx.io/docs/tokenomics/gmx-token) into buying GMX back on the open market. Per the docs, those bought-back tokens accumulate in the treasury and staking-reward distribution is currently paused, set to resume once GMX reaches a target price — a governance choice to convert ongoing revenue into buy pressure and a treasury reserve instead of a running dividend.

## The July 2025 V1 exploit

On 9 July 2025, the legacy **GMX V1** GLP pool on Arbitrum was drained of roughly [$40–42 million](https://www.theblock.co/post/362164/gmx-token-surges-14-after-hacker-begins-returning-funds-from-40-million-exploit). The attack exploited a re-entrancy flaw that let the attacker mint GLP abnormally: by opening manipulated short positions they distorted the pool's recorded average short price and size, inflated the GLP token price, and redeemed at the inflated value. GMX halted V1 trading and GLP minting, and the team offered the attacker a [10% white-hat bounty](https://www.dlnews.com/articles/defi/gmx-offers-hacker-10-percent-bounty-after-42m-exploit/) to return the rest — which the exploiter [accepted and repaid](https://cointelegraph.com/news/gmx-exploiter-begins-returning-stolen-funds). The episode is the sharpest illustration of the pooled model's core trade-off: when LPs _are_ the counterparty, a pricing bug isn't a bad trade for one desk — it's a direct claim on the shared pool everyone funded. V2's isolated markets and stricter oracle handling are the structural response.

## How Caper approaches this

GMX ties governance power to a freely-tradeable, capped token: buy and stake more GMX and you earn a larger share of fees and a larger vote over the parameters that set those fees. It is transparent capital-weighting — whoever brings the most liquidity-underwriting capital carries the most weight. [Caper](/wiki/foundations/what-is-a-caper) weights a member's say differently. Governance weight is the canonical formula `w = (t · v) / (V · T)` — a member's held tokens `t` multiplied by the vote tokens `v` they have actually earned, over total vote supply `V` and circulating tokens `T` (verified against `compute_vote_weight` in `contracts/common/src/lib.rs`, the shared crate the logic and ballot packages both fold weights with). The vote count `v` is _earned_ on both of a caper's surfaces — minted one per ballot cast and 0.01 per XRD of gross trade value — as a non-transferable, fractional proof-of-vote token no one can sell you (verified in `contracts/core/src/caper_dao.rs`), and the same weight sets a member's [exit](/wiki/foundations/leaving-a-caper) share of the treasury. Holdings still matter — `t` is a multiplier, so a member with no stake has no weight — but capital cannot buy another member's accrued `v` the way stacking GMX can. The design goal is different: GMX optimizes for whoever underwrites the most risk, Caper for whoever has both skin in the game and a record of showing up. See [Trading](/wiki/markets/trading) and [Voting](/wiki/governance/voting) for the mechanics.

## References

- [GMX — official site](https://gmx.io/)
- [GMX Docs — GMX token, supply, staking & buyback](https://docs.gmx.io/docs/tokenomics/gmx-token)
- [GMX Docs — Providing liquidity (GM pools, fee split)](https://docs.gmx.io/docs/providing-liquidity/)
- [Arbitrum blog — GMX in depth](https://blog.arbitrum.io/gmx-an-in-depth-look-at-arbitrums-leading-permissionless-exchange-for-on-chain-leverage-trading/)
- [GMX governance — Snapshot](https://snapshot.org/#/gmx.eth) · [Tally](https://www.tally.xyz/gov/gmx)
- [The Block — July 2025 GMX V1 exploit & fund return](https://www.theblock.co/post/362164/gmx-token-surges-14-after-hacker-begins-returning-funds-from-40-million-exploit)
- [Cointelegraph — exploiter returns stolen funds](https://cointelegraph.com/news/gmx-exploiter-begins-returning-stolen-funds)