That's already kind of how tools like Opencode Claude Code and Copilot work now. Copilot doesn't do it unless you remind the model is has the power to use a subagent, but yeah it works very well for that exact purpose.
Some people use it to save tokens or time for work. Like use opus for planning, then fire off some gpt or sonnet, or smaller models for other tasks. I often use the free models to do things like read and summarize code changes then report back to the main model. Copilot currently doesn't charge for submodel usage, but im sure they will in time.
It's also possible in Opencode to mix in your own local models too, although I've been struggling with it for a while.
I just setup some automated PR tasks using opencode. I tired using my self hosted models, but they couldn't fight their way out of a wet paper bag unfortunatly. I spent way too much time playing with it for now.
ChipTuner
npub1qd…3fqm7
2026-02-07 20:20:34 UTC
in reply to nevent1q…mj64
Author Public Key
npub1qdjn8j4gwgmkj3k5un775nq6q3q7mguv5tvajstmkdsqdja2havq03fqm7Published at
2026-02-07 20:20:34 UTCEvent JSON
{
"id": "14d99d385824d510bddd746493824129ddc65cd179fb9d96335bd04e59e6d265",
"pubkey": "036533caa872376946d4e4fdea4c1a0441eda38ca2d9d9417bb36006cbaabf58",
"created_at": 1770495634,
"kind": 1,
"tags": [
[
"e",
"2f3700fe68d2629c0b9dc413746bf772b5b8bc2f6001b5b62d9711606a299ffd",
"wss://theforest.nostr1.com/",
"root",
"44dc1c2db9c3fbd7bee9257eceb52be3cf8c40baf7b63f46e56b58a131c74f0b"
],
[
"e",
"2f3700fe68d2629c0b9dc413746bf772b5b8bc2f6001b5b62d9711606a299ffd",
"wss://theforest.nostr1.com/",
"reply",
"44dc1c2db9c3fbd7bee9257eceb52be3cf8c40baf7b63f46e56b58a131c74f0b"
],
[
"p",
"44dc1c2db9c3fbd7bee9257eceb52be3cf8c40baf7b63f46e56b58a131c74f0b"
]
],
"content": "That's already kind of how tools like Opencode Claude Code and Copilot work now. Copilot doesn't do it unless you remind the model is has the power to use a subagent, but yeah it works very well for that exact purpose. \n\nSome people use it to save tokens or time for work. Like use opus for planning, then fire off some gpt or sonnet, or smaller models for other tasks. I often use the free models to do things like read and summarize code changes then report back to the main model. Copilot currently doesn't charge for submodel usage, but im sure they will in time. \n\nIt's also possible in Opencode to mix in your own local models too, although I've been struggling with it for a while. \n\nI just setup some automated PR tasks using opencode. I tired using my self hosted models, but they couldn't fight their way out of a wet paper bag unfortunatly. I spent way too much time playing with it for now. ",
"sig": "10b06f0f129677a19fcd562f3a0641cfae29839121bf42c6de4a270c6579417c9960daa13427da5ea74c0f49c3e0370739bfba504bb48c5c8da37a1828bd95f1"
}