what is the most compatible llm to be used with opencode? #3180

Open
opened 2026-02-16 17:39:01 -05:00 by yindo · 4 comments
Owner

Originally created by @aivagrin on GitHub (Nov 28, 2025).

Question

hitting limit always for context length in new session when switching etc 137k for deepseek etc.

how to automatically trim / compress command for previous context input or how to work around this?

Originally created by @aivagrin on GitHub (Nov 28, 2025). ### Question hitting limit always for context length in new session when switching etc 137k for deepseek etc. how to automatically trim / compress command for previous context input or how to work around this?
Author
Owner

@github-actions[bot] commented on GitHub (Nov 28, 2025):

This issue might be a duplicate of existing issues. Please check:

  • #4845: prompt is too long unrecoverable - Directly addresses context length limits with Anthropic models and mentions needing to compact
  • #2464: AT_APICallError: prompt token count exceeds the limit - Similar token limit issue blocking workflow
  • #2841: Context length issue / better context inspector - Questions about hidden context usage and tracking
  • #4317: Feature: generic /compact command, auto-compaction, and fork-aware conversations - Proposes a comprehensive solution for automatic context trimming/compression
  • #1212: Fetched documentation exceeds context window limit - Automatic truncation possible - Early request for automatic truncation of context
  • #3153: Feature: Subagents should support auto compaction - Related feature request for auto-compaction

These issues cover both the problem you're experiencing (hitting context limits with DeepSeek) and several proposed solutions (auto-compaction, context compression). Feel free to ignore if your specific case involves DeepSeek compatibility concerns not addressed by these.

@github-actions[bot] commented on GitHub (Nov 28, 2025): This issue might be a duplicate of existing issues. Please check: - #4845: prompt is too long unrecoverable - Directly addresses context length limits with Anthropic models and mentions needing to compact - #2464: AT_APICallError: prompt token count exceeds the limit - Similar token limit issue blocking workflow - #2841: Context length issue / better context inspector - Questions about hidden context usage and tracking - #4317: Feature: generic /compact command, auto-compaction, and fork-aware conversations - Proposes a comprehensive solution for automatic context trimming/compression - #1212: Fetched documentation exceeds context window limit - Automatic truncation possible - Early request for automatic truncation of context - #3153: Feature: Subagents should support auto compaction - Related feature request for auto-compaction These issues cover both the problem you're experiencing (hitting context limits with DeepSeek) and several proposed solutions (auto-compaction, context compression). Feel free to ignore if your specific case involves DeepSeek compatibility concerns not addressed by these.
Author
Owner

@rekram1-node commented on GitHub (Nov 30, 2025):

@aivagrin are you defining deepseek in your own opencode.json? there are a lot of things that could cause this

Here are some models we think you'll like:
https://opencode.ai/docs/models/#recommended-models

This ins't complete, there are a lot of options to choose from

@rekram1-node commented on GitHub (Nov 30, 2025): @aivagrin are you defining deepseek in your own opencode.json? there are a lot of things that could cause this Here are some models we think you'll like: https://opencode.ai/docs/models/#recommended-models This ins't complete, there are a lot of options to choose from
Author
Owner

@Cyrille37 commented on GitHub (Dec 3, 2025):

Hi
I'm joining this discussion, I hope I'm not disturbing anyone ;-)

About compatibility between AI coding tool :

I'm running llama.cpp on a RTX3060 12Go with "Qwen2.5-coder-7b-instruct-q8_0" model.
When asking to "write method's comments in @"

  • cline well do the job (it does too much like git commit, I have to add rules...)
  • opencode reply "Understood. bla bla" or it writes a new Unit test, but never add comment to methods

Is because of system prompt ? Other tips ?

Thanks & regards

@Cyrille37 commented on GitHub (Dec 3, 2025): Hi I'm joining this discussion, I hope I'm not disturbing anyone ;-) About compatibility between AI coding tool : I'm running llama.cpp on a RTX3060 12Go with "`Qwen2.5-coder-7b-instruct-q8_0`" model. When asking to "write method's comments in @<a unit-test file>" - [`cline`](https://cline.bot/) well do the job (_it does too much like git commit, I have to add rules..._) - `opencode` reply "Understood. bla bla" or it writes a new Unit test, but never add comment to methods Is because of [system prompt](https://github.com/sst/opencode/tree/dev/packages/opencode/src/session/prompt) ? Other tips ? Thanks & regards
Author
Owner

@rekram1-node commented on GitHub (Dec 3, 2025):

I would need to see the session to see what went wrong

@rekram1-node commented on GitHub (Dec 3, 2025): I would need to see the session to see what went wrong
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#3180