GPT models 1M context by dagmfactory · Pull Request #11838 · warpdotdev/warp

dagmfactory · 2026-05-28T16:45:00Z

Description

Enabling 1M Context for supported GPT models (gpt-5.4, gpt-5.5, grape)
OpenAI specific long-context pricing warning when context window >272K (OpenAI threshold)
Server as source of truth for which models support configurable context windows
change is behind GPTConfigurableContextWindow flag
removing stale flag configurable_context_window

Testing

I have manually tested my changes locally with ./script/run

Screenshots / Videos

Default

Warning

Agent Mode

Warp Agent Mode - This PR was created via Warp's AI Agent Mode

…arpdotdev/warp into dagm/allow-gpt-models-1M-context

…ls-1M-context

oz-for-oss · 2026-05-28T20:06:10Z

@dagmfactory

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

dagmfactory · 2026-05-28T20:06:54Z

                base: params.model.into(),
                cli_agent: params.cli_agent_model.into(),
                computer_use_agent: params.computer_use_model.into(),
-                base_model_context_window_limit: if FeatureFlag::ConfigurableContextWindow


@coolcom200 same question here.
Is it okay to remove this feature flag now?

Yes should be safe to remove.

oz-for-oss

Overview

This PR replaces the old configurable context window gate with GPT-specific OpenAI gating, relies on server-provided model metadata for context-window support, adds long-context pricing warnings in the relevant settings/editor surfaces, and covers the new gating/sanitization behavior with unit tests. No approved spec context was available, and the supplemental security pass did not find security concerns.

Concerns

One non-blocking UI consistency issue: the warning path should use the same clamped effective limit as the request path so stale persisted values do not show an OpenAI pricing warning when the model's current max is below the warning threshold.

Verdict

Found: 0 critical, 0 important, 1 suggestions

Approve with nits

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss · 2026-05-28T20:10:27Z

+            Some(
+                self.context_window_limit
+                    .unwrap_or(llm.context_window.default_max),
+            ),


💡 [SUGGESTION] Clamp the warning input the same way context_window_limit_for_request does. A stale persisted override above the current model max would make this warning appear even though the outgoing request and slider are constrained to the lowered range.

Suggested change

Some(

self.context_window_limit

.unwrap_or(llm.context_window.default_max),

),

Some(

self.context_window_limit

.unwrap_or(llm.context_window.default_max)

.clamp(llm.context_window.min, llm.context_window.max),

),

Co-Authored-By: Oz <oz-agent@warp.dev>

oz-for-oss · 2026-06-02T17:38:57Z

@dagmfactory

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss

Overview

This PR replaces the old broad configurable-context-window flag with GPT-specific gating, routes request context-window limits through model metadata sanitization, and adds OpenAI long-context pricing warnings in the AI settings/profile UI. It also updates the reusable warning box to support formatted hyperlink titles and adds unit coverage for the new context-window helper behavior.

Concerns

No blocking correctness, security, or spec-alignment concerns found in the annotated diff.

Verdict

Found: 0 critical, 0 important, 0 suggestions

Approve

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

…arpdotdev/warp into dagm/allow-gpt-models-1M-context # Conflicts: # app/src/view_components/warning_box.rs

danielpeng2

FYI the screenshots in the description aren't loading.

I tried using long context with your server change and I keep getting errors?

Also a product nit: can we make the warning in the settings appear as you drag, not only after you release the slider?

danielpeng2 · 2026-06-02T19:28:46Z

+    /// Enables configurable expanded context windows for eligible GPT models.
+    GPTConfigurableContextWindow,


ooc do we need a client-side feature flag for this? the server already drives whether long context for GPT models is enabled -- it feels sufficient to let the server handle that decision

I think the server should be the source of truth for which models are and are not configurable. We can add a flag on the client side just as a control for rollout and clean it up later though.

we can already control the rollout via the server, and avoiding the client flag seems like it would make some of the client code simpler.

not a huge deal though, up to you!

dagmfactory · 2026-06-02T20:50:00Z

can we make the warning in the settings appear as you drag, not only after you release the slider

Can you send me the conversation ID? I can take a look. It's working on my end with the server changes https://www.loom.com/share/491850b7038241519ff7a45612af5e5c

danielpeng2 · 2026-06-03T04:46:38Z

@@ -303,13 +303,7 @@ impl RequestParams {
                .active_profile(terminal_view_id, app)
                .data()
                .clone();


do we still need the clone here if we're going to call context_window_limit_for_request to get the limit directly? feels like we could simplify this part of the code now

danielpeng2 · 2026-06-03T04:53:13Z

+    }
+}
+
+pub(crate) fn sanitize_context_window_limit_for_request(


this method seems like it's only used in context_window_limit_for_request? I have a slight preference for inlining the method, feels like it'd be a bit easier to read.

danielpeng2 · 2026-06-03T05:21:42Z

+    pub fn with_formatted_title(mut self, formatted_title: FormattedTextInline) -> Self {
+        self.formatted_title = Some(formatted_title);
+        self
+    }


the API is a bit awkward here since to construct a WarningBoxConfig with a formatted title, you have to go through the new constructor and pass in a non-formatted title that will get overwritten.

we have a couple other UI components where the label can be either text or some custom element. a good example of this is RadioButtonItem. it's a bit cleaner since we can use an enum to encode the "text or custom element" constraint in code.

danielpeng2 · 2026-06-03T05:34:14Z

+
+    selected_limit.map(|limit| limit.clamp(llm.context_window.min, llm.context_window.max))
+}
+pub(crate) fn has_effective_configurable_context_window(


nit: what does it mean semantically to have an "effective" configurable context window. could we name the method has_effective_configurable_context_window?

danielpeng2 · 2026-06-03T05:45:30Z

 use crate::server::sync_queue::QueueItem;
 use crate::settings::AISettings;
 use crate::workspaces::user_workspaces::UserWorkspaces;
+pub const LONG_CONTEXT_WARNING_THRESHOLD: u32 = 272_000;


we can do this in another PR, but I think we should pass the threshold value down from the server. we might need to modify the graphql/protos to pass that info alongside the rest of the context window values.

having this be server driven makes it easier to support new models with different thresholds in the future. imagine OpenAI releases a new model with a different threshold value. we want to avoid needing an out-of-band client release just to support the new model.

can we specify in a comment that this threshold is only relevant for GPT 5.4 + 5.5 for now, and wire the threshold value down from the server in a follow-up PR?

danielpeng2 · 2026-06-03T05:48:22Z

+    /// Enables configurable expanded context windows for eligible GPT models.
+    GPTConfigurableContextWindow,


we can already control the rollout via the server, and avoiding the client flag seems like it would make some of the client code simpler.

not a huge deal though, up to you!

danielpeng2 · 2026-06-03T05:49:35Z

can we make the warning in the settings appear as you drag, not only after you release the slider

Can you send me the conversation ID? I can take a look. It's working on my end with the server changes https://www.loom.com/share/491850b7038241519ff7a45612af5e5c

I tried again on your updated server branch, and it worked fine. Might've had something weird in my local setup, ignore that error!

dagmfactory added 7 commits May 27, 2026 11:17

Allow configurable context window for GPT models

d328f47

feature flag

fc16b71

Refactor context window handling + tests

3fc3f82

add to dogfoodfeature flag

1296477

allow byok context window setting + tests

62f3f23

remove duplicate logic for cinfigurable window eligibility

fd4e1b4

remove duplicate logic for configurable window eligibility

0b316fd

cla-bot Bot added the cla-signed label May 28, 2026

dagmfactory added 2 commits May 28, 2026 15:05

Merge branch 'dagm/allow-gpt-models-1M-context' of ssh://github.com/w…

8667597

…arpdotdev/warp into dagm/allow-gpt-models-1M-context

Merge remote-tracking branch 'origin/master' into dagm/allow-gpt-mode…

da70e2f

…ls-1M-context

dagmfactory marked this pull request as ready for review May 28, 2026 20:05

dagmfactory commented May 28, 2026

View reviewed changes

oz-for-oss Bot reviewed May 28, 2026

View reviewed changes

formatting

7733dfd

dagmfactory marked this pull request as draft May 28, 2026 22:28

dagmfactory and others added 11 commits June 1, 2026 11:39

learn more hyperlink

c07bb42

Use Shrinkable for sliders in AI editor and settings page

ad98a0c

Allow configurable context window for GPT models

290564a

feature flag

877e619

Refactor context window handling + tests

636abbe

add to dogfoodfeature flag

9248f3c

allow byok context window setting + tests

4a8fdc0

remove duplicate logic for configurable window eligibility

a4c913f

learn more hyperlink

aee457e

Use Shrinkable for sliders in AI editor and settings page

37abf7b

Merge origin/master into dagm/allow-gpt-models-1M-context

e501393

Co-Authored-By: Oz <oz-agent@warp.dev>

dagmfactory marked this pull request as ready for review June 2, 2026 17:38

oz-for-oss Bot reviewed Jun 2, 2026

View reviewed changes

dagmfactory requested a review from danielpeng2 June 2, 2026 17:49

dagmfactory added 3 commits June 2, 2026 14:14

fmt

9a68abd

Merge branch 'dagm/allow-gpt-models-1M-context' of ssh://github.com/w…

ac8fd55

…arpdotdev/warp into dagm/allow-gpt-models-1M-context # Conflicts: # app/src/view_components/warning_box.rs

Merge branch 'master' into dagm/allow-gpt-models-1M-context

83bc513

danielpeng2 reviewed Jun 2, 2026

View reviewed changes

warning shows while dragging

974f66d

fmt

2b3f782

danielpeng2 reviewed Jun 3, 2026

View reviewed changes

		/// Enables configurable expanded context windows for eligible GPT models.
		GPTConfigurableContextWindow,

Conversation

dagmfactory commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Testing

Screenshots / Videos

Default

Warning

Agent Mode

Uh oh!

oz-for-oss Bot commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Verdict

Uh oh!

oz-for-oss Bot May 28, 2026

Choose a reason for hiding this comment

Uh oh!

oz-for-oss Bot commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Verdict

Uh oh!

danielpeng2 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dagmfactory commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

danielpeng2 commented Jun 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

dagmfactory commented May 28, 2026 •

edited

Loading

oz-for-oss Bot commented May 28, 2026 •

edited

Loading

oz-for-oss Bot commented Jun 2, 2026 •

edited

Loading

dagmfactory commented Jun 2, 2026 •

edited

Loading