ElizaOS vs Virtuals: When the Framework Wins and the Token Doesn't

The asymmetry, in one paragraph

ElizaOS built the open-source AI agent codebase that earned developer mindshare in web3, then bolted on a token nobody can value. Virtuals Protocol built a token model with mandatory base-pair demand and real fee capture, then built it on a centralised inference engine that breaks the moment Virtuals’ servers go down. Same agent category. Opposite trade-offs. Both tokens are down roughly 90% from their peaks. The reason that asymmetry keeps showing up is the actual story.

I hold a small position in VIRTUAL. I don’t hold ELIZAOS. Both project reviews are linked above with full scoring; this article is about the gap between the two models, not the scores in isolation.

The codebase: Eliza wins; Virtuals isn’t really competing here

ElizaOS is the most popular open-source AI agent codebase in web3 by GitHub metrics: 17,662 stars, 583 contributors, 5,439 forks, 17,545 commits, 90+ official plugins, MIT licensed. You install it via npm or bun, define a character file, swap the LLMLLMLarge Language Model. A neural network trained on vast amounts of text to predict the next word in a sequence. Modern LLMs (GPT, Claude, Llama, Qwen, DeepSeek) generate human-quality text and are the foundation of most modern AI products.Like an autocomplete that read every book ever written. It has no memory of individual texts but it has absorbed the patterns of language so deeply that it can generate paragraphs that sound human. The skill is statistical, not conscious.Read more → backend (OpenAI, Anthropic, Google, or Ollama for local inferenceInferenceRunning a trained AI model to produce an answer. Inference is what happens when you type a prompt into ChatGPT and get a response. The model takes your input, computes a best guess, and returns it.Like asking an expert for their opinion. The training was the decades they spent becoming an expert. The inference is the 30 seconds it takes them to answer your specific question.Read more →), and connect it to Discord, Telegram, X, or Farcaster. Three minutes from install to a running agent, give or take.

Virtuals’ equivalent is the GAME SDK, which is open source and reasonably documented. But the GAME engine itself, the part that runs inference for every agent on the platform, is hosted on Virtuals’ own servers. There is no permissionless path to running GAME inference. If Virtuals’ infrastructure goes down, every agent on the platform stops talking. That isn’t an open-source codebase you can compare to ElizaOS on the same axis. It’s a hosted product with an SDK on top.

The wider context matters. LangChain, the agent library most ElizaOS contributors are competing with in their day jobs, has 118,000+ GitHub stars and no token. CrewAI, AutoGPT and the rest of the open-source agent stack have no tokens either. Open-source agent code tends to win on adoption, and tokens aren’t usually how it gets there. That’s the structural problem ElizaOS is trying to solve in reverse.

The token model: Virtuals wins; Eliza isn’t really competing here

VIRTUAL is the mandatory base pair for every agent token launched on the platform. Every agent token on Base, Solana, Ronin, Arbitrum or XRP Ledger pairs against VIRTUAL in its initial liquidity pool. Every trade routes through it. Agent creation costs 100 VIRTUAL. veVIRTUAL holders vote, accrue Virgen points and qualify for Genesis airdrops. The Unicorn launch system uses an anti-sniper tax that starts at 99% and decays to 1% over 98 minutes. None of that is theoretical: it’s running across five chains, with 18,000+ agents launched and roughly $39.5M cumulative protocol revenue (per Virtuals’ own reporting, partially verifiable on-chain via fee contracts).

ELIZAOS’s token utility is a sentence with a lot of qualifiers. There is no stakingStakingLocking up a cryptocurrency to help secure a blockchain network, usually in exchange for rewards. The locked tokens act as a security deposit that can be taken away if the staker misbehaves.Like putting down a large rental deposit for an apartment. You get the money back if you behave, you earn interest while it's locked, and the landlord takes it if you trash the place.Read more →. No fee accrual. No burnBurnPermanently removing tokens from circulation by sending them to an address that no one controls. Burns reduce total supply, which (all else equal) makes each remaining token worth more of the network's value.Like a company buying back its own shares and shredding them. The company's total value stays the same, but each remaining share now represents a slightly bigger slice of that value.Read more → mechanism. The governance voting module has been promised since launch and isn’t live; in practice, decisions get made by Eliza Labs and signal-boosted on Discord. The most promising piece, an L1L1Layer 1. A base blockchain that runs its own consensus mechanism, executes transactions, and settles its own state. Bitcoin, Ethereum, NEAR, and Solana are all L1s. Anything built on top of an L1 is technically a Layer 2 or higher.Like the foundation of a building. Nothing else can exist on top until the foundation is solid. Different L1s make different tradeoffs for what kind of building they can support.Read more → blockchain with ELIZAOS as native gas, has been in discussion with Saga since early 2025 with no concrete implementation. The token sits next to the codebase, not inside it. You can use ElizaOS itself without ever touching the token, which is what most of its 583 contributors do.

Fact: ELIZAOS scores 8/20 on Token Utility and 3/20 on Value Accrual in the OYM dual-score system. VIRTUAL scores 16/20 and 14/20 on the same dimensions.

Take: that’s not a small gap. It’s the difference between a token that has reasons to exist and one that doesn’t, and the prices reflect it.

Distribution and dilution: opposite stories

ElizaOS started with one of the cleaner launch profiles in the agent space. Fair launchFair LaunchA token launch where everyone has the same access from day one. No private sale, no insider allocation, no VC discount. Tokens are distributed by mining, staking, or open public sale at a single price.Like a 100m sprint where everyone starts behind the same line at the same time. Some runners are faster, but nobody gets to start 10 metres ahead because they paid extra. The race is decided by the run, not by who bought the best position.Read more → on DAOs.fun in October 2024. 420.69 SOL raised. No pre-mine, no VCVCVenture Capital. Private investors who fund projects at an early stage in exchange for equity or token allocations. VC rounds are typically pre-launch, at steep discounts to any future public price, with multi-year vesting.Like angel investors in a startup who buy shares before the company goes public. They take more risk because the company might fail, so they get a better price. Once the company IPOs they can sell, and the public market pays whatever price it thinks is fair.Read more → allocation, no team allocation. The original AI16Z token was as community-first as the agent category produced.

Then in November 2025, Eliza Labs migrated AI16Z to ELIZAOS at a 1:6 swap ratio and expanded total supply by 40%, from 6.6 billion to 11 billion tokens. The new tokens funded allocations that didn’t exist at the original launch: 15% to SAFT investors, 10% to the team, 4.5% to a foundation, 2.5% to a growth fund. Insider allocation post-migration sits at roughly 32%. Original community holders kept 60% at the swap rate; the rest came from issuing new tokens against them. VestingVestingA schedule that locks up tokens allocated to insiders, investors, and team members, releasing them gradually over months or years. Vesting prevents insiders from dumping on public buyers immediately after launch.Like a new employee's stock options at a startup. You don't get all the shares on day one. They unlock over four years so you stick around and do the work rather than cashing out and leaving.Read more → is enforced on-chain via Streamflow contracts, which is better than handshake vesting. The 40% expansion itself was decided unilaterally without an on-chain governance vote, because on-chain governance doesn’t exist.

Virtuals went the other way. PathDAO raised $16.61M entirely through public IDOs in December 2021 (Fjord Foundry LBP, Enjinstarter, PAID Network), then migrated PATH to VIRTUAL in late 2023. 60% of supply in public circulation, 35% in a DAO-controlled treasury (capped at 10% annual emission for three years), 5% in liquidity. No VC allocation. No team allocation. All tokens are fully unlocked, which means there are no future cliffCliffA waiting period at the start of a token vesting schedule during which no tokens unlock at all. After the cliff ends, tokens begin releasing according to the vesting schedule.Like a probationary period at a new job. You don't get your stock options on day one. You wait 12 months to prove you'll stick around, then everything starts unlocking normally.Read more → events to worry about and no insider lockup expiries hanging over the price.

That’s a serious distribution by current standards. The 35% DAO-controlled treasury is the real concentration concern, and the team retains practical influence over how it’s deployed, but at the level of who got the tokens at launch, Virtuals has fewer skeletons. Eliza’s distribution looks worse the longer you stare at the migration, not better.

Revenue: one of these projects has any

Virtuals captures fee revenue at multiple layers. The 1% trading fee on agent tokens splits between protocol and agent SubDAOs. Agent creation costs 100 VIRTUAL per launch. Inference payments to GAME route through VIRTUAL. At peak in January 2025, daily protocol revenue exceeded $1M. Cumulative revenue across the protocol’s history sits at roughly $39.5M, per the project’s own reporting (the underlying fee contracts on Base are verifiable on-chain). Roughly 13 million VIRTUAL has been routed to a buyback-and-burn programme, taking 25 agent tokens permanently out of supply. The Revenue Network announced at Consensus Hong Kong on 12 February 2026 distributes up to $1M per month specifically to revenue-generating agents, an attempt to push the platform’s incentives toward agents that earn rather than agents that just trade.

ElizaOS has none of this. The codebase is open-source and free to use, which means there’s nothing to charge for at the open-source layer. auto.fun, the no-code agent launchpad, generates some fees, but the link between auto.fun fees and ELIZAOS token value is undefined. The original “AI VC fund” thesis, where the agent “Marc AIndreessen” autonomously runs a treasury, has no published portfolio performance. The DAO treasury value is unverified.

The honest counter on Virtuals: protocol revenue collapsed by 97% in the two months after the January 2025 peak. Multi-chain expansion has broadened the fee surface, but it hasn’t proved cycle-independent yet. Revenue is real. Whether it’s durable through a downturn is the open question.

Both have governance problems, in different shapes

This is where the two projects start to look more alike, even though the failure modes differ.

Virtuals has functional on-chain governance. veVIRTUAL holders can create proposals at 0.10% of supply, vote during a 72-hour window, and pass at 25% quorum and simple majority. The mechanism works. The catch is that the team retains substantial influence through the 35% treasury, and a January 2025 smart-contract vulnerability disclosure was initially ignored, with the bug-bounty Discord channel closed before the team patched the issue under public pressure. Smart contracts on Base are upgradeable. So the machinery exists; a single team controls enough levers that the machinery’s outputs are predictable.

ElizaOS has no functional governance machinery at all. The DAO voting module is roadmap. Major decisions, including the rebranding from AI16Z, the ticker change, the 1:6 swap ratio, the SAFT round, and the 40% supply expansion, were all made by Eliza Labs without an on-chain vote. The AI agent “Marc AIndreessen” retains final authority over treasury investment decisions. Token holders can suggest things in Discord. They cannot bind anyone to anything.

Different failure modes, similar outcome: in both cases, a small team makes the calls that matter, and token holders ratify after the fact.

What would each need to score highly?

Both projects sit in the agent quadrant of the OYM dual-score system. Neither is in Quadrant A. Here’s what would actually move them.

Eliza’s path

1. Ship binding on-chain governance. Token holders need to vote, and votes need to bind. The 40% supply expansion would not have passed an on-chain vote at the prices token holders paid before the migration. That’s exactly why it should have been one.
2. Give the token a reason to exist. Either ship the L1 with ELIZAOS as native gas, or formalise auto.fun fee accrual to holders, or run a buy-and-burn funded by platform revenue. The current state, where you can use the codebase for free and ignore the token, is sustainable for the open-source product and fatal for the token.
3. Patch the Princeton memory injection vulnerability. Researchers demonstrated that attackers can poison agent memory via social media interactions; Eliza Labs’ public response was “no direct updates were made.” For agents that handle financial transactions, that’s not a position you can hold while asking institutions to take you seriously.

Ship all three and ELIZAOS plausibly clears 60 on Returns and 65 on Freedom, within range of Quadrant A. Ship none and the token stays a speculative sidecar to a good open-source product. The open question is whether Eliza Labs prioritises the token or stays focused on the codebase, where most of their real traction lives.

Virtuals’ path

1. Open-source or decentralise the GAME engine. This is the single biggest freedom deficit on the platform. As long as inference for every agent runs on Virtuals’ servers, the team is the platform and the token is a derivative on a hosted product. A permissionless inference network, or a hybrid where third parties run GAME nodes, would transform the centralisation profile.
2. Demonstrate cycle-independent revenue. The $39.5M cumulative is real. The 97% peak-to-trough collapse is also real. The Revenue Network and the Eastworld Labs robotics vertical are bets on broadening revenue beyond agent-launchpad fees. If the next downturn doesn’t produce another 97% collapse, the case for VIRTUAL gets fundamentally stronger. If it does, the buyback-and-burn runs on diminishing fuel.
3. Move agent data to decentralised storage. Personality, memory, voice models and visual assets all live on Virtuals’ servers. “Owning” an agent currently means owning a token that points at IP held by a platform. Migrating that data to IPFS, Arweave or similar would give creators real ownership, which is the actual sovereignty pitch.

Ship all three and VIRTUAL clears 65 on Freedom while holding above 65 on Returns. The realistic trajectory, given the team’s iteration cadence, is incremental: Unicorn launch system, buyback-and-burn, Revenue Network, multi-chain expansion. None of those yet addresses the centralised inference engine, which is the load-bearing one.

The structural pattern: why this trade-off keeps appearing

The agent-token category produces the same shape of trade-off over and over. Open-source frameworks earn developer mindshare and don’t need tokens to do it; the LangChain pattern wins on adoption. Tokenised platforms earn fee revenue and need centralised infrastructure to function, because permissionless inference at production latency isn’t really solved yet. Pick one side or the other and you get half the picture.

Morpheus is the closest project to a third path: a permissionless compute marketplace that pays for inference through a token-native mechanism, with the work itself done by independent providers. It isn’t in the agent category exactly, but it’s the architecture the agent category would need to borrow from to escape the trade-off. Until that matures, or until Virtuals or a competitor ships something equivalent, the agent-token space will keep producing pairs like ElizaOS and Virtuals: one project with the code but no token thesis, one with the token thesis but no real decentralisation, both losing market value in lockstep with whatever narrative cycle is currently driving fee revenue.

Same category, opposite trade-offs, both down roughly 90% from peak. The asymmetry isn’t a quirk of these two projects; it’s the shape of the agent-token space until something changes.