Fix bedrock max tokens

[MuriloFP]

Related GitHub Issue

Closes: #2500

Description

Fix: Update AWS Bedrock Claude model max output tokens to match official specs

Changes:

• Updated maxTokens values in packages/types/src/providers/bedrock.ts to align with official Anthropic documentation

Models Updated:

• Claude Sonnet 4 (anthropic.claude-sonnet-4-20250514-v1:0): 8,192 → 64,000 tokens
• Claude Opus 4 (anthropic.claude-opus-4-20250514-v1:0): 8,192 → 32,000 tokens
• Claude 3.7 Sonnet (anthropic.claude-3-7-sonnet-20250219-v1:0): 8,192 → 64,000 tokens

Impact:

• Users now see correct max output token limits in settings UI
• Enables generation of significantly longer responses (up to 7.8x increase)
• Resolves discrepancy between displayed limits and AWS Bedrock documentation

References:

• AWS Bedrock Claude 3.7 documentation: 64k tokens (128k with beta header)
• Anthropic official model comparison table

Test Procedure

• Open Settings, select Amazon Bedrock as a provider
• Select any of the affected models
• Look at the Max Tokens value

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Documentation Impact: I have considered if my changes require documentation updates (see "Documentation Updates" section below).
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Screenshots / Videos

Documentation Updates

Additional Notes

I've looked over Bedrock official documentation, and it links back to Anthropic's. I don't believe there are different values for models used through Bedrock. There is no explicit list of the mentioned models with their Max Output values in Bedrock's documentation, only on Anthropic's.

Get in Touch

@MuriloFP

---

Important

Update maxTokens for three models in bedrock.ts to align with official documentation, correcting UI limits and resolving documentation discrepancies.

• Behavior:
  • Updated maxTokens in bedrock.ts for models anthropic.claude-sonnet-4-20250514-v1:0, anthropic.claude-opus-4-20250514-v1:0, and anthropic.claude-3-7-sonnet-20250219-v1:0.
  • New maxTokens values: 64,000, 32,000, and 64,000 respectively.
• Impact:
  • Corrects max output token limits in settings UI.
  • Allows generation of longer responses, up to 7.8x increase.
  • Fixes discrepancy with AWS Bedrock documentation.

^{This description was created by for 6d29f44. You can customize this summary. It will automatically update as commits are pushed.}

[mrubens]

What does the output tokens default to after this change? Do these have sliders?

[MuriloFP]

What does the output tokens default to after this change? Do these have sliders?

For most users: They'll now get the full model capability by default (64k tokens for Claude 3.7 Sonnet, 32k for Opus 4) with no sliders needed - its just set at the correct limit.
For reasoning-enabled models: Defaults to 16,384 tokens but sliders are available in the "Thinking Budget" section to adjust up to the model's full capability.
Bottom line: Regular users get an improvement (8k → 64k tokens) automatically, while users with reasoning enabled still have full control via sliders.

No other defaults were changed, just the model information that was incorrect.

[daniel-lxs]

With this change the defaults look like this for me:

[mrubens]

What does the output tokens default to after this change? Do these have sliders?

For most users: They'll now get the full model capability by default (64k tokens for Claude 3.7 Sonnet, 32k for Opus 4) with no sliders needed - its just set at the correct limit. For reasoning-enabled models: Defaults to 16,384 tokens but sliders are available in the "Thinking Budget" section to adjust up to the model's full capability. Bottom line: Regular users get an improvement (8k → 64k tokens) automatically, while users with reasoning enabled still have full control via sliders.

No other defaults were changed, just the model information that was incorrect.

My concern is that I don't actually think that 64k output tokens is clearly an improvement. That 64k is always reserved in the context window, reducing the amount that's usable for input. I think it makes sense that the user can opt into it, but I think the default shouldn't be higher than 16k for modes with a 200k context window.

[hannesrudolph]

Closing as we need to setup a max token slider across providers where models have the ability to have variable tokens set.