Morpheus bids quote pricePerSecond in MOR. Most upstream APIs quote per-token or per-month. To set a defensible bid, you need to translate upstream cost into per-second MOR.

​
A simple model

For a per-token upstream: 
upstream_$_per_second ≈ tokens_per_second_average × upstream_$_per_token_blended
                      = tps × ((p × in_$/tok) + ((1−p) × out_$/tok))
Where p is the share of input tokens (often ~0.3 for chat). Convert dollars to MOR using the current MOR/USD rate. For a flat-rate (subscription) upstream: 
upstream_$_per_second ≈ subscription_$_per_month / (utilized_seconds_per_month)
You should pricePerSecond ≥ upstream_$_per_second + your_margin + risk_buffer.

​
Floor and minimums

ItemValue
bidPricePerSecondMin10000000000 wei (0.00000001 MOR/sec)
Marketplace bid fee (paid once)0.3 MOR
Provider stake (refundable)0.2 MOR (or 10000 MOR for subnet)
Model stake (refundable)0.1 MOR
You cannot post below the floor. There is no upper limit, but consumer rating algorithms will skip you if you’re far above other providers serving the same model.

​
A worked example

Suppose Venice Diem costs you $30/mo and you expect ~100k seconds of utilized inference per month: 
$/sec = 30 / 100000  ≈  0.0003 USD/sec
If 1 MOR ≈ $X (check live), then MOR/sec ≈ 0.0003 / X. Add a ~30% margin for headroom.

​
Iterating on pricing

  • Watch your bid acceptance rate. If you get no sessions, you may be priced too high.
  • Watch upstream margin. If sessions cost more upstream than they bring in MOR, raise pricing or reduce concurrency.
  • Look at competing bids on active.mor.org.

​
Updating an existing bid

Bids are immutable on chain — to change pricing, delete the old bid (DELETE /blockchain/bids/:id) and post a new one. Plan for the bid fee on each post.

​
Risk hygiene

  • Keep a small ETH balance for gas; running out mid-session is bad UX.
  • Keep extra MOR allowance approved to the Diamond contract so re-posting bids doesn’t fail mid-month.
  • For multi-model resale, post one bid per modelId — the proxy-router routes by modelId deterministically.