Support xAI Grok 4.6 on AWS Bedrock

Ideas Feature Requests
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Grok 4.6 went live on Bedrock on 2026-08-19 (inference profiles us.xai.grok-4.6, global.xai.grok-4.6). Adding either ID as a Bedrock model in Cursor and sending a message fails with:
“We encountered an issue when using your API key: Selected model is not supported by bedrock, please use a different model”

The same account invokes the model fine directly (aws bedrock-runtime converse --model-id us.xai.grok-4.6 … succeeds, including tool use), so this is Cursor’s Bedrock model allowlist, not an AWS problem. Client logs show resolvedParams=0 for these IDs vs 1 for cataloged models.

Steps to Reproduce

  1. Configure AWS Bedrock as a model provider in Cursor Settings → Models (access key/secret). Anthropic Bedrock models (e.g. us.anthropic.claude-opus-5) work normally in this setup.
  2. In the model list, add a custom model with the ID of xAI Grok 4.6’s Bedrock inference profile: us.xai.grok-4.6. Grok 4.6 has been live on Bedrock since 2026-08-19.
  3. Select the model in chat and send any message.

Result: the request fails with:
“We encountered an issue when using your API key: Selected model is not supported by bedrock, please use a different model”

The same AWS account invokes the model successfully outside Cursor, including tool use:
aws bedrock-runtime converse --model-id us.xai.grok-4.6
–messages ‘[{“role”:“user”,“content”:[{“text”:“Say ok”}]}]’
–inference-config ‘{“maxTokens”:100}’
So the rejection is Cursor’s Bedrock model allowlist, not an AWS/IAM/model-access problem. Client renderer.log shows resolvedParams=0 for these IDs vs resolvedParams=1 for cataloged Bedrock models.

Expected Behavior

Grok 4.6’s Bedrock inference profiles should be routable like other Bedrock models.

Two implementation notes from direct API testing:

  • The model rejects the temperature and topP fields with a ValidationException (“This model doesn’t support the temperature field”); requests must send maxTokens only.
  • Responses lead with reasoningContent.redactedContent blocks (encrypted reasoning), which the response parser needs to accept.

Operating System

MacOS

Version Information

Version: 3.14.27
VS Code Extension API: 1.128.0
Commit: 047548b00c1a079373d74d00183f32510a4a41e0
Date: 2026-08-04T03:29:11.183Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.291
OS: Darwin arm64 23.1.0

For AI issues: which model did you use?

us.xai.grok-4.6 and global.xai.grok-4.6 (AWS Bedrock, custom model ID)

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

7

Hey @Aden_Clemente
BYOK Bedrock currently routes Anthropic Claude models only, so us.xai.grok-4.6 and global.xai.grok-4.6 get rejected before the request reaches AWS (expected today, nothing wrong with your setup). In the meantime Grok 4.6 is available directly in Cursor’s model picker without an API key. We’ve let the team know and it’s an issue we’re tracking - thanks for the notes on the sampling params and reasoning blocks.

8

Thanks @mohitjain – is this solely on Cursor’s end or would this be limitation with any usage of Bedrock?