Repository navigation
[Bug] Custom provider models fail with "maxOutputTokens must be >= 1" when limit is not defined #22253
Closed as not planned
Description
Description
After updating to v1.4.3, custom provider models fail with:
Invalid argument for parameter maxOutputTokens: maxOutputTokens must be >= 1
Root Cause
There is an asymmetric fallback between built-in models and custom models (found via binary string analysis):
- Built-in models:
output: limits?.output ?? 64000→ safe fallback to 64000 - Custom models:
output: model.limit?.output ?? existingModel?.limit?.output ?? 0→ falls back to 0, causing the error
When a custom provider model has no limit field defined, maxOutputTokens becomes 0, which is rejected by the AI SDK validation.
Reproduction
Define a custom provider model without a limit field in opencode.json:
"models": {
"claude-sonnet-4-6": {
"name": "Claude Sonnet 4.6",
"type": "chat",
"available": true
}
}Then run opencode with this model — error occurs immediately on any prompt.
Workaround
Add limit to each model in the custom provider config:
"claude-sonnet-4-6": {
"name": "Claude Sonnet 4.6",
"type": "chat",
"available": true,
"limit": { "context": 200000, "output": 64000 }
}Expected Behavior
Custom models without a limit field should fall back to a reasonable default (e.g. 64000) instead of 0, consistent with built-in model behavior.
Version
opencode v1.4.3
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Apr 13, 2026
This issue might be a duplicate of existing issues. Please check:
maxOutputTokens must be >= 1) with the same root cause — a custom/local provider model without alimitfield defined, causing the output tokens to fall back to 0. That issue reports this via Ollama, while this one is via a manually configured custom provider, but they appear to be the same underlying bug.MertSoylu commentedon Apr 13, 2026
I've fixed this issue! The problem was in
provider.tswhere custom models without alimitfield were falling back to 0 instead of OUTPUT_TOKEN_MAX. Changed the fallback to use ProviderTransform.OUTPUT_TOKEN_MAX (32000), consistent with built-in models.PR: #22291
wjiuxing commentedon Apr 14, 2026
@comzip I have also added similar limit information to the model, but I still encounter this problem.
comzip commentedon Apr 15, 2026
@wjiuxing I added it to opencode.json as shown below and it works normally.
"models": { "claude-sonnet-4-6": { "name": "Claude Sonnet 4.6", "type": "chat", "available": true, "limit": { "context": 200000, "output": 64000 } }wjiuxing commentedon Apr 15, 2026
@comzip Yes. It's working well in a new session but failed in the old interrupted one. Thanks a lot.
github-actions commentedon Jul 5, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.