[BUG]: GPT 5.2 context_length_exceeded / Your input exceeds the context window of this model.

[peixotorms]

Description

Using GPT 5.2 via api, with reasoning high.

After about 1 hour of usage, I hit

{"type":"error","sequence_number":2,"error":{"type":"invalid_request_error","code":"context_length_exceeded","message":"Your input exceeds the context window of this model. Please adjust your input and try again.","param":"input"}}

On the right panel I see

Context
270,593 tokens
68% used
$5.13 spent

I noticed that context keeps going up, then down, which from what I read, its intended.
However, maybe its not counted correctly, because GPT 5.2 limit via api is 400k not 270k.

/compact works, then I can proceed, so this is just a report on the incorrect count and limit.

Thanks

OpenCode version

v1.0.193

Steps to reproduce

Just use normally with gpt 5.2 high.

Screenshot and/or share link

No response

Operating System

Ubuntu Server

Terminal

No response

[github-actions]

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

• Research subagent exceeded the context window and silently failed #5970: Research subagent exceeded the context window and silently failed
• [FEATURE] Sliding window context management for long-running sessions #4659: Sliding window context management for long-running sessions (addresses context overflow scenarios)
• Model in BUILD mode does not have enough context to continue after compaction #3031: Model in BUILD mode does not have enough context to continue after compaction
• GPT5 incorrect context counting #1714: GPT5 incorrect context counting (similar context counting discrepancy)
• prompt is too long unrecoverable #4845: prompt is too long unrecoverable (similar unrecoverable context limit errors)

The core issue appears to be context limit counting accuracy and handling for GPT 5.2. Feel free to ignore if none of these address your specific case.

I am also encountering this issue.

[clayrisser]

I am facing this issue with custom local models. Is there a way to configure the context size for local models?

[rekram1-node]

@clayrisser make sure you have the context input & output defined for your model in the model definition

[Tioit-Wang]

GPT-5.2 xhigh

OpenCode Desktop
v1.1.53

[rafaell-lycan]

Same here

Context
238,028 tokens
60% used
$8.72 spent

Also using GPT-5.2 (xhigh)

OpenCode Desktop:
v1.1.56

@thdxr any ideas?

[peixotorms]

Well, after this I stumble across some articles about the 272k limit, so the 400k OpenAI advertises as context is 272k in + 128k out.

Sneaky as it sounds, they changed the description to "context" for marketing purposes, starting on GPT 5.
In reality, its 272k input + 128k output, not "context" as we know it, and think that it's just input.

Still there is a bug here, to count it correctly.

[mvarchdev]

Have this also, hope it gets fixed, as I quite like opencode - but for now, switching back to codex app