[BUG] [Regression][Desktop] inferenceModels[].prefer1m no longer defaults the 1M-context variant in the Cowork model picker (3p / gateway inference)

Status Closed — duplicate
Reported on v2.1.217
Maintainer reply None cached
Activity 3 comments · opened Jul 23, 2026 · closed Aug 25, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

What's Wrong?

In Cowork on third-party (gateway) inference, setting prefer1m: true on the default (first) inferenceModels entry does not make the 1M-context variant the default selection in the model picker. The picker opens on the standard (200K) variant instead.

supports1m itself works correctly — the "Opus 4.8 (1M context)" entry is offered in the picker and is selectable. Only the defaulting behavior described for prefer1m in the configuration reference fails to take effect:
_Default to 1M context (prefer1m) makes the 1M-context variant the default picker selection when this entry is the default model (the first entry)._

This is distinct from #69444 (where supports1m was dropped for 3p and the variant disappeared entirely — that is fixed for us; the variant appears). Here the variant appears but is never promoted to the default.

This is a regression. With an unchanged bootstrap config, prefer1m correctly defaulted the picker to the 1M variant in an app version from roughly in early July. A subsequent update (current: 1.24012.1) broke the defaulting behavior — the 1M variant is still offered, but is no longer selected by default.

What Should Happen?

On first run with no prior explicit pick, the default-selected model should be Opus 4.8 (1M context) — the 1M variant of the first/default entry, as prefer1m specifies.

Error Messages/Logs

Steps to Reproduce

Configure Cowork for gateway/3p inference via bootstrap-server mode (GET /user/bootstrap returning JSON config; Claude Desktop ≥ 1.10270.0).
Return an inferenceModels list whose first entry has supports1m: true and prefer1m: true and is the family default, e.g.:
json
"inferenceModels": [
{"name": "claude-opus-4-8", "labelOverride": "Opus 4.8",
"supports1m": true, "prefer1m": true,
"anthropicFamilyTier": "opus", "isFamilyDefault": true},
{"name": "claude-sonnet-5", "labelOverride": "Sonnet 5",
"anthropicFamilyTier": "sonnet", "isFamilyDefault": true},
{"name": "claude-haiku-4-5", "labelOverride": "Haiku 4.5",
"anthropicFamilyTier": "haiku", "isFamilyDefault": true}
]

(prefer1m: true is confirmed present in the JSON served on the wire.) 3. Sign in on a fresh profile / machine with no prior model selection (to rule out a persisted "explicit user pick" overriding the default). 4. Open the Cowork model picker and observe the default-selected model.

Claude Model

Other

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.217

Platform

AWS Bedrock

Operating System

Windows

Terminal/Shell

Other

Additional Information

App: Claude Desktop 1.24012.1
Last known good: an app version from ~mid-to-late June 2026 (same bootstrap config, prefer1m defaulted correctly) — regression introduced in an update since then
OS: Windows
Surface: Cowork tab
Inference: third-party gateway (bootstrap-server mode)

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗