TL;DR: OP_RETURN is cheaper for data smaller than 143 bytes.
The overhead of an OP_RETURN output is at least 11 bytes:
- amount (8 bytes)
- compact size of script (at least 1 byte)
- script:
- OP_RETURN opcode (1 byte)
- push opcode (at least 1 byte)
The overhead of an inscription is at least 118.75 vB (84 base bytes and 139 witness bytes):
- P2TR output (43 bytes)
- P2TR input:
- outpoint TXID (32 bytes)
- outpoint index (4 bytes)
- sequence (4 bytes)
- compact size of input script (1 byte, script is empty)
- witness data:
- compact size of witness stack (1 WU)
- compact size of signature (1 WU)
- signature (64 WU)
- compact size of script (at least 1 WU)
- script:
- public key push opcode (1 WU)
- public key (32 WU)
- OP_CHECKSIG opcode (1 WU)
- OP_FALSE opcode (1 WU)
- OP_IF opcode (1 WU)
- push opcodes (at least 1 WU)
- OP_ENDIF opcode (1 WU)
- compact size of control block (1 WU)
- control block (33 WU)
The witness discount cancels out the larger overhead at 143 bytes of data, which takes 155 vB using an OP_RETURN and 154.75 vB using an inscription. For anything smaller than that OP_RETURN is cheaper.
Notes:
- The above assumes a best-case scenario in which the P2TR input is included in a transaction that serves some other purpose and would be made anyway. This often isn’t the case, in which case the overhead is at least 129.25 vB and OP_RETURN is cheaper for up to at least 158 bytes of data.
- It’s necessary to split the data into chunks of 520 bytes in an inscription but not in an OP_RETURN output. This only affects data large enough that it doesn’t change the answer.
- We’re ignoring the 80-byte standardness limit for OP_RETURN here.
