What is the specific context size of Github Copilot, is it only 128K?

Question

As the title says, I'm considering purchasing GitHub Copilot.

May I ask if GitHub Copilot's Claude model only has a 128K context size (whether used in VSCode or OpenCode)?

If it only has a 128K context, does that mean the model's capabilities are severely limited?

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

Is there a way to increase Context limit for Github Copilot Provider? #5993: Same question about GitHub Copilot context limit being capped at 128K tokens in OpenCode
Github Copliot provider allows only 128k token conversations. #2317: Earlier report of GitHub Copilot provider only allowing 128K token conversations (closed with discussion)
github copilot: 2x token (counts total=total+cached) compacts at 64k, instead of 128k #3314: Related bug about token counting causing compaction to trigger early at 64K instead of 128K

These issues cover the same underlying question about GitHub Copilot's context window limitations in OpenCode.

Github copilot has custom limits for different models, usually 128k but not always. 4o only has 64k, and some codex models get 272k, this is basically only visible in the model picker in the VScode extension of Copilot
This does impact the model, but less than you would think.
Usually the performance degrades rapidly after about 100k - 120k context used. I have both an anthropic key and Copilot, and using Opus through both I find that I usually compact / restart sessions before 128k context is filled anyway

@jeroenev

@jeroenev
I don't have a Copilot subscription, so I can only test the following two models. Haiku should have 200K, and GPT-5-mini should have 272K, but when I run /context in the Copilot CLI, both only show 160K. Could it be that they increased the context size? Is the maximum now 160K? Can you test what the total context size shows when running /context with Sonnet or Opus?

Seems like they increased the context recently, just 2 weeks ago this was 128k for most models:

@jeroenev

@jeroenev May I ask which software this screenshot is from or where you saw it? Thank you.

That is in VS-code, in the Agent Chat panel from Github Copilot
Click the model selector -> More models -> Manage models

@jeroenev

@jeroenev Thank you, this makes it more transparent. It seems the 1M model won't be available for Copilot pricing in the near future.

Seems like they increased the context recently, just 2 weeks ago this was 128k for most models:

Will the context size in OpenCode be increased as well? I' m using OMO plugin, the auto compaction will be triggered frequently if the task is complex with 128K context size.

Similar problem on OpenCode failing to "phone home" to check provider context size #16709

If you don't feel like waiting keep in mind the context size limit can be adjusted in opencode.json:

  "provider": {
    "github-copilot": {
      "models": {
        "claude-haiku-4.5": {
          "limit": { "context": 160000, "output": 32000 }
        },
        "claude-opus-4.5": {
          "limit": { "context": 160000, "output": 32000 }
        },
        "claude-opus-4.6": {
          "limit": { "context": 192000, "output": 64000 }
        },
        "claude-sonnet-4": {
          "limit": { "context": 144000, "output": 16000 }
        },
        "claude-sonnet-4.5": {
          "limit": { "context": 160000, "output": 32000 }
        },
        "claude-sonnet-4.6": {
          "limit": { "context": 160000, "output": 32000 }
        },
        "gemini-2.5-pro": {
          "limit": { "context": 173000, "output": 64000 }
        },
        "gemini-3-flash-preview": {
          "limit": { "context": 173000, "output": 64000 }
        },
        "gemini-3-pro-preview": {
          "limit": { "context": 173000, "output": 64000 }
        },
        "gemini-3.1-pro-preview": {
          "limit": { "context": 173000, "output": 64000 }
        },
        "gpt-4.1": {
          "limit": { "context": 128000, "output": 16384 }
        },
        "gpt-4o": {
          "limit": { "context": 68000, "output": 16384 }
        },
        "gpt-5-mini": {
          "limit": { "context": 192000, "output": 64000 }
        },
        "gpt-5.1": {
          "limit": { "context": 192000, "output": 64000 }
        },
        "gpt-5.1-codex": {
          "limit": { "context": 256000, "output": 128000 }
        },
        "gpt-5.1-codex-max": {
          "limit": { "context": 256000, "output": 128000 }
        },
        "gpt-5.1-codex-mini": {
          "limit": { "context": 256000, "output": 128000 }
        },
        "gpt-5.2": {
          "limit": { "context": 192000, "output": 64000 }
        },
        "gpt-5.2-codex": {
          "limit": { "context": 400000, "output": 128000 }
        },
        "grok-code-fast-1": {
          "limit": { "context": 173000, "output": 64000 }
        }
      }
    },

I give no guarantees this will work correctly and that the values are right, I just mapped what I saw in vscode and I'm dogfooding it right now.

@d4rky-pl

@d4rky-pl Looking forward to the results of your test.

@tlerbao

@tlerbao works for me with Opus 4.6. I don't know if/when I'll check other models and this is definitely temporary config until models.dev gets updated

@d4rky-pl

@d4rky-pl Where can I receive the official updates about this configuration as soon as possible?

Any fix sirs? this should be

@SadE54 Why doesn't my copilot cli have 200K

Cause it seems the way opencode is trying to get some metadata from Copilot concening models is outdated . I guess it will be fixed soon . But you can use an already existing plugin to fix this by yourself. Just add in you opencode.json file : "plugin": [ "@zhzy0077/opencode-copilot-cli-auth@0.0.16" ]

i did but still getting only 128k on Opus 4.6? already removed and added provider again

Any fix sirs? this should be

@SadE54 Why doesn't my copilot cli have 200K

Cause it seems the way opencode is trying to get some metadata from Copilot concening models is outdated . I guess it will be fixed soon . But you can use an already existing plugin to fix this by yourself. Just add in you opencode.json file : "plugin": [ "@zhzy0077/opencode-copilot-cli-auth@0.0.16" ]

i did but still getting only 128k on Opus 4.6? already removed and added provider again

Strange , it's working for me , for opus 4.6

btw , it seems there's some commits added today for this problem

With 1.3 it seems I got a ~140K for Claude models , still not the 200k 🤔

With 1.3 it seems I got a ~140K for Claude models , still not the 200k 🤔

The latest API results seem to indicate that Sonnet is 200k, while Opus is only 144k.

With 1.3 it seems I got a ~140K for Claude models , still not the 200k 🤔

The latest API results seem to indicate that Sonnet is 200k, while Opus is only 144k.

Yeah , I checked with Sonnet , and got the same. it's kind a progress 😅

With 1.3 it seems I got a ~140K for Claude models , still not the 200k 🤔

The latest API results seem to indicate that Sonnet is 200k, while Opus is only 144k.

Yeah , I checked with Sonnet , and got the same. it's kind a progress 😅

But it seems everyone prefers using opus more, haha.

The API results are weird compared to what vscode displays:

 "family": "claude-opus-4.6",
  "limits": {
    "max_context_window_tokens": 144000,
    "max_non_streaming_output_tokens": 16000,
    "max_output_tokens": 64000,
    "max_prompt_tokens": 128000,

"family": "claude-sonnet-4.6",
"limits": {
  "max_context_window_tokens": 200000,
  "max_non_streaming_output_tokens": 16000,
  "max_output_tokens": 32000,
  "max_prompt_tokens": 128000,

vs

@d4rky-pl

@d4rky-pl @SadE54
Yes, the API and VS Code display differently.

What type of subscription do you have? Could it be related to the subscription type?

@tlerbao

@tlerbao individual Pro+

Here's the API results for business Pro:

 "family": "claude-opus-4.6",
  "limits": {
    "max_context_window_tokens": 200000,
    "max_non_streaming_output_tokens": 16000,
    "max_output_tokens": 32000,
    "max_prompt_tokens": 168000,

  "family": "claude-sonnet-4.6",
  "limits": {
    "max_context_window_tokens": 200000,
    "max_non_streaming_output_tokens": 16000,
    "max_output_tokens": 32000,
    "max_prompt_tokens": 168000,

FYI @rekram1-node ^

I would be inclined to say to use the @zhzy0077/opencode-copilot-cli-auth if it actually pulls the values from API but the plugin is currently broken (subagents fail with invalid initiator error) so I'd rather see this officially supported by opencode at some point (and I'll just continue running with my custom overrides in opencode.json)

@d4rky-pl

@d4rky-pl
Thank you so much! I tried @zhzy0077/opencode-copilot-cli-auth myself, but for the opus-4.6 model, the usage percentage shown in the top-right corner during runtime calculates back to a limit of 144k. This is consistent with the update on models.dev (discussed in anomalyco/models.dev#1136). So what should I do now? Should I configure it like this in opencode.json?

"claude-opus-4.6": { "limit": { "context": 192000, "output": 64000 } }

But you mentioned that the API results for business pro show "max_context_window_tokens": 200000 and "max_output_tokens": 32000.

Also, thank you so much for pointing out that the plugin is currently broken (subagents fail with invalid initiator error). I was assigning tasks to the subagent, and it kept showing "delegating." I searched through the issues for a long time without finding the reason, and now I finally know why.

Im confused but is there a fix implemented already for this?

GH Copilot Context Size update models.dev#1136 (comment)

I'm still a little confused.

we sync directly from copilot apis in opencode and fallback to models.dev so the context limits match what they let u access fro their plan

Navigation Menu

Repository navigation

What is the specific context size of Github Copilot, is it only 128K? #16129

Question

12 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

What is the specific context size of Github Copilot, is it only 128K? #16129

Description

Question

Activity

github-actions commented on Mar 5, 2026

jeroenev commented on Mar 5, 2026

tlerbao commented on Mar 5, 2026

jeroenev commented on Mar 5, 2026

tlerbao commented on Mar 6, 2026

jeroenev commented on Mar 6, 2026

tlerbao commented on Mar 6, 2026

hjnnjh commented on Mar 8, 2026

TomLucidor commented on Mar 9, 2026

d4rky-pl commented on Mar 10, 2026

tlerbao commented on Mar 10, 2026

d4rky-pl commented on Mar 10, 2026

tlerbao commented on Mar 10, 2026

12 remaining items

krasen1234 commented on Mar 17, 2026

SadE54 commented on Mar 17, 2026

SadE54 commented on Mar 23, 2026

tlerbao commented on Mar 23, 2026

SadE54 commented on Mar 23, 2026

tlerbao commented on Mar 23, 2026

d4rky-pl commented on Mar 23, 2026

tlerbao commented on Mar 23, 2026

d4rky-pl commented on Mar 23, 2026

enterlet commented on Mar 26, 2026

krasen1234 commented on Mar 27, 2026

jensenojs commented on Mar 28, 2026

rekram1-node commented on May 13, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions