Live Inference Activity

Two metrics, two trust models

MOR earned is the bottom-line economic activity on the network, calculated trustlessly from session start, end, and price. Tokens processed is provider-attested: the input + output counts signed into the closeout receipt. Treat one as audited revenue, the other as a self-reported throughput figure.

Why the two rows can move apart

MOR earned is priced on time rather than tokens. The on-chain formula is (min(closedAt, endsAt) − openedAt) × bid.pricePerSecond, so a provider earns on how long a session stays open and the bid it was opened at, whatever the throughput inside that window. Tokens served counts the work itself. The two are independent, so MOR earned can spike while tokens served holds at its usual level.

When that happens, the driver is longer sessions or higher bids, more committed compute time or pricier capacity, rather than more inference. The tell is MOR per session and MOR per token both jumping while the session count stays flat or falls. Read that as provider revenue and reserved capacity moving, not a demand surge.

For a read on actual usage, the throughput row is the cleaner signal. Attested tokens served is the inference delivered, independent of price and duration, with session count and the spread of models served as corroboration. MOR earned answers how much providers were paid; tokens served answers how much the protocol was used.

Where the MOR actually comes from

Over 98% of all sessions are opened with isDirectPaymentFromUser = false: the user locks collateral that gets returned to them at session close, and the provider is paid out of the protocol's fundingAccount treasury on Base. The treasury is funded by the 24% compute pool slice of MOR emissions. Most inference today is subsidised by emissions, not paid for by users. The runway implications are spelled out next.

The direct-payment path barely exists in practice. Just 4,823 sessions have ever used it, 1.5% of the total, all between 8 March and 27 April 2026 across three wallets, and they paid providers 65 MOR between them. Nothing since. Session length is the reason: the contract computes it from your stake and the compute-pool ratio without reading the direct-payment flag at all, so paying buys exactly the compute staking would have given you free. No interface exposes the option either, and the Morpheus session-lifecycle docs document only the staked flow.

The gateway's dollar side sits outside all of this. Card and crypto payments settle off-chain with whoever runs the gateway, so any decision to buy and stake MOR with that revenue is the operator's own, and the protocol mandates nothing of the sort. Treat gateway fiat as off-chain operator revenue that the protocol neither sees nor depends on, and look for structural MOR demand only where the chain can actually show it.

Sustainability headroom

Read against the Diamond on 15 August 2026, the on-chain compute pool (getComputeBalance) holds 2,799,521 MOR. The contract caps daily settlement at 1% of that, so the budget is 27,995 MOR per day. Provider claims run at roughly 800 MOR per day, about 2.9% of the cap.

The pool itself tells a two-phase story. The compute pillar accrues about 2906 MOR per day. Through July claims ran well ahead of that and the balance fell from a peak of 2,809,457 MOR on 20 July to 2,753,562 by 1 August, about 4,650 MOR a day. Almost all of it was one wallet exploiting a stake-lock bug that a facet swap closed on 31 July. Since then the pool has been growing again, currently by 2106 MOR per day, and the funding Safe that actually pays providers holds 115,952 MOR, roughly 145 days of claims. Any rate quoted for this network has to say whether its window straddles 31 July 2026, because a 30-day average across that date still reports the pool draining.

The access side grew alongside it. MOR committed to inference access, counting the gateway's opener wallet with everything held in the Diamond as session stake and holds, went from 123,259 MOR on 16 June 2026 to 801,079 MOR today. The gateway sits well below its ceiling, though: across the first fortnight of August it drew about 33% of the daily ration its own committed MOR entitles it to, and 3.5% on the last day of that window. So the position grew ahead of demand rather than because the ration was binding. Access demand is visible on-chain, but users posting their own MOR is token demand rather than revenue.

A caveat about the early window

The Morpheus API ran an open beta from contract redeploy in December 2025 until 2 March 2026, when it went officially live. Across the beta, providers quoted near-zero prices by design, so MOR earned is flat across the entire pre-launch period regardless of session volume. Some of the provider-signed token counts from the beta are also clearly test traffic. Headline numbers above sum the last 60 days, which falls almost exactly inside the post-launch era and keeps them representative of paid-era activity. Lifetime totals across the full dataset (including beta) live in the underlying JSON.

Provider concentration

Thirty-one addresses have ever closed a session and recent days run with seven to thirteen, so the supply side has widened since this page first went up. The concentration moved to value instead. One operator, mordiem, runs two registered providers: mordiem.com, which opens its own sessions, and router.mordiem.com:3333, which serves the gateway. Together they account for 79% of every MOR the network has ever paid a provider. That is a recent shift: mordiem earned nothing before May 2026 and mor.org's own provider led the network through March and April. Provider earnings are capped at each provider's posted stake per 365-day period, so this concentration tracks committed capital rather than being unbounded. Pre-redeploy activity at the prior contract address is out of scope for this MVP.

Stake here doesn't pay providers; it just measures your share of the daily credit allocation. The MOR that ends up paying providers comes from the protocol treasury, as covered above. So this chart and the inference activity chart together describe the full economic picture: how much MOR is parked for access vs how much is being settled per session. Both are tracked daily.

Sources and reproducibility

• Inference Contract (LumerinDiamond): 0x6aBE1d282f72B474E54527D93b979A4f64d3030a on Base mainnet
• Source code: MorpheusAIs/Morpheus-Lumerin-Node
• Pipelines: scripts/refresh-mor-inference.ts and scripts/refresh-mor-subnet-staking.ts in the OYM repo. Reproducible against any Alchemy Base RPC endpoint.
• Refresh cadence: daily via GitHub Actions. Last refreshed: 2026-08-25.

If you want to use this dataset in your own research or writing, please attribute Own Your Mind and link back to this page.