Slack Cloud Agent blocks GPT-5.6 Sol when only its 1M context option is disabled

Support Bug Reports
, , ,
1

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Launching a Cloud Agent from Slack with the default model gpt-5.6-sol-high fails before creating a bc- agent.

GPT-5.6 Sol is enabled in our team model settings. Only its 1M context option is disabled. @Cursor settings still displays gpt-5.6-sol-high, but launching produces either:

“The agent failed to start. Please try again.”

or:

“This model is blocked by your team’s model settings.”

Grok 4.6 launches successfully from the same Slack installation and repository.

This looks like Slack may be silently resolving the saved Sol default to context=1m, while its settings card displays only the model and effort level. The UI provides no way to see or select the context value.

Steps to Reproduce

  1. Enable GPT-5.6 Sol in Team Settings → Models.
  2. Disable only the Sol 1M context option.
  3. Configure the Slack Cloud Agent default as GPT-5.6 Sol High.
  4. Run @Cursor settings and confirm it displays gpt-5.6-sol-high.
  5. Mention @Cursor with any prompt that should start an agent.
  6. Observe that the launch fails before a bc- agent is created.
  7. Select Grok 4.6 instead and repeat. The agent launches successfully.

The Sol launch failed repeatedly.

Request ID:
serverGenReqId_ba55078b-d672-4e38-9f6a-885cd5d783dc

Approximate failure time:
September 22, 2026, 12:58 UTC

Expected Behavior

Slack should launch GPT-5.6 Sol using its documented 272k default context when 1M is unavailable.

If Slack requires 1M, the settings UI should expose that context selection and explain that the 1M parameter, rather than the entire Sol model, is blocked.

Operating System

MacOS

Version Information

N/A — this is a server-side Slack Cloud Agent launch failure. Cursor IDE was not involved. Slack and the Cursor dashboard were accessed through Chrome on macOS.

For AI issues: which model did you use?

GPT-5.6 Sol High (gpt-5.6-sol-high), non-Fast. Slack does not display the selected context-window parameter.

For AI issues: add Request ID with privacy disabled

serverGenReqId_ba55078b-d672-4e38-9f6a-885cd5d783dc

The request failed before Cursor created a bc- Cloud Agent ID.

Additional Information

GPT-5.6 Sol is enabled in our team model policy; only its 1M context option is disabled. Sol’s documented default context is 272k.

The same Slack installation, account, repository, and base branch successfully launch Grok 4.6 agents. This isolates the problem to the effective Sol model selection or its policy validation.

Please inspect the stored Slack defaultModelSelection, resolved model.params, and effective team-policy decision for the request ID above. In particular, check whether Slack attached context=1m while displaying only gpt-5.6-sol-high.

Does this stop you from using Cursor

No - Cursor works, but with this issue

4

Update: the issue is currently resolved after refreshing the model defaults.

Before retesting, I:

  1. Changed Reasoning from High to Medium, then back to High.
  2. Disabled “Use Team Default,” then re-enabled it.

I did not change model access or the 272K context setting.

Slack then successfully launched GPT-5.6 Sol as agent bc-d1b22a0d-7a11-5a40-9ae2-b753a90ee0a1.

Because both settings were refreshed before the successful launch, I cannot isolate which action fixed it. The behavior suggests that Slack was using a stale or incorrectly persisted effective model selection. Rewriting the model parameters and re-inheriting the team default cleared the incorrect “blocked by your team’s model settings” error.

11

Hey @tobias.counsel, glad it’s launching again, and thanks for the update.

Your fix is real but the cause is a little different from a stale setting. The Slack settings card and the team-level default only store the model name, and Slack runs a name-only default in its largest context window (1M for GPT-5.6 Sol), which your team has disabled. When you re-saved the model from the cursor.com dashboard, the saved default included the 272k context window, and that is why the next launch went through.

Two things to watch for:

  1. The agent that launched (bc-d1b22a0d-7a11-5a40-9ae2-b753a90ee0a1) is running Sol with Medium reasoning, not High. If you want High, set it from the dashboard’s Cloud Agent default model picker with the 272k context selected, rather than from the Slack card.

  2. Picking the model again from the Slack settings card, or clicking “Use Team Default” (the team default is still saved in the name-only form), will bring the error back. Until this is fixed, treat the dashboard as the place to set the default. A team admin re-saving the team default from the dashboard will fix that side too.

I’ll flag this for the team!

I’ve flagged this to the team so Slack falls back to 272k when 1M isn’t allowed and the message explains what is actually blocked.

16

This is broader than Slack. I get the same error when I start a Sol Cloud Agent directly from Cursor Web.

Our team allows GPT-5.6 Sol at 272K and blocks the 1M option. Cursor exposes both context choices, and I explicitly selected 272K.

Cursor 2026-09-23 11.24.21
Cursor 2026-09-23 11.24.211314×492 20.9 KB

This makes the team setting unusable. We want people to use Sol at 272K without accidentally starting long-context runs. Cursor charges 2x for input and 1.5x for output once a Sol request exceeds 272K, so this restriction is an intentional cost control because we’ve seen threads that accidentally spent severral hundreds of dollars repeatedly with Sol.

The current behavior gives us two choices. We can enable 1M for everyone just to make Sol work in the cloud, or we can make Sol unavailable to Cloud Agent users. Neither is acceptable. About 90% of my Cursor work happens in Cloud Agents, so using Sol only in local sessions is not a practical workaround.

Existing Sol Cloud Agents can still process follow-up messages. Only new runs fail. This suggests that the launch path resolves the model name to 1M and ignores the selected 272K context.

A Cloud Agent configured for 272K should launch at 272K. A model selection without an explicit context should use the context allowed by team policy. Cursor should not silently resolve it to a blocked 1M configuration.

Please treat this as a Cloud Agent model-selection and team-policy bug, not a Slack-specific issue.

21

Hey @tobias.counsel, thanks for the follow-up, and you’re right that this isn’t specific to Slack!

The screenshot is from the Cloud launcher inside the Cursor desktop app. That launcher currently starts new Cloud Agents in the largest context mode no matter which context you pick, so on a team that restricts 1M the launch gets rejected. The team is working on a change so a lower context pick in the desktop app launches at that lower context.

In the meantime, starting the agent from cursor.com/agents in your browser with GPT-5.6 Sol and 272K selected should launch normally. Slack also works as long as the default was saved from the dashboard with 272K selected.

I’ve added your report to the issue we’re tracking, including your point that a selection without an explicit context should use what your team allows.