Repository navigation
[FEATURE]: Fill context limits for OpenAI-compatible providers from their /models advertisement #53235
Description
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
Custom @ai-sdk/openai-compatible providers are not in models.dev. Without a hand-written limit block for every model, OpenCode has no context window for them and auto-compaction cannot know when to fire. Long sessions grow until the upstream rejects them with context_length_exceeded.
Most OpenAI-compatible servers (vLLM, LM Studio, LiteLLM, Ollama, OpenRouter) already publish per-model context_length / max_input_tokens / max_output_tokens on GET {baseURL}/models. OpenCode could fill the missing limits from there, with a config limit still winning.
Reproduction
Configure an openai-compatible provider with one model and no limit, point it at a server that advertises limits, run a long session. Compaction never triggers and the upstream eventually 400s. Earlier 2.x releases defaulted unknown models to a 200k context, which compacted at the wrong point for 1M-window models.
System details
- OpenCode: 2.0.22 (desktop bundle CLI); verified against dev @ 907b3bc
- OS/client: macOS 15, desktop app and CLI sessions
- Provider: self-hosted OpenAI-compatible proxy (OmniRoute) serving 1M-context GLM 5.3 and 262k-context DeepSeek models; applies equally to vLLM / LM Studio / LiteLLM / Ollama endpoints
PR: #53236
Activity
github-actions commentedon Oct 4, 2026
This issue does not fully meet our contributing guidelines.
What needs to be fixed:
/models), so please use the Feature Request template: start the title with[FEATURE]:and include the required verification checkbox.Please edit this issue to address the above within 72 hours, or it will be automatically closed.
This issue might be a duplicate of existing issues. Please check:
/models./modelsmetadata.If you believe this was flagged incorrectly, please let a maintainer know.
[-]OpenAI-compatible providers: advertised context limits on /models are ignored, so auto-compaction never fires[/-][+][FEATURE]: Fill context limits for OpenAI-compatible providers from their /models advertisement[/+]github-actions commentedon Oct 4, 2026
Thanks for updating the issue with the Feature Request details, reproduction steps, and system information!
Tarquinen commentedon Oct 20, 2026
We're tracking this in #40908, so I'm closing this as a duplicate. Please follow that issue for updates.
— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊