Integrate grok-4-fast-non-reasoning for xai provider

[juliettech13]

• Integrate grok-4-fast-non-reasoning for xAI provider
• Updated information regarding the grok-4-fast-reasoning
• Tests & snapshots

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
helicone	Ready	Preview	Comment	Oct 8, 2025 9:52pm
helicone-bifrost	Ready	Preview	Comment	Oct 8, 2025 9:52pm
helicone-eu	Ready	Preview	Comment	Oct 8, 2025 9:52pm

[claude]

Claude finished @juliettech13's task —— View job

---

Comprehensive Code Review: xAI Grok-4-Fast-Non-Reasoning Integration

Score: 7/10 - Solid implementation with minor improvements needed

Issues and Recommendations

1. Model Name Inconsistency - Medium Priority

Location: packages/cost/models/authors/xai/models.ts:38

// Current inconsistent naming
"grok-4-fast-reasoning": { name: "xAI: Grok 4 Fast Reasoning" }
"grok-4-fast-non-reasoning": { name: "xAI Grok 4 Fast Non-Reasoning" } // Missing colon

Fix: Standardize naming format for consistency:

name: "xAI: Grok 4 Fast Non-Reasoning"

2. Audio Modality Configuration Gap - Medium Priority

Location: packages/cost/models/authors/xai/models.ts:45 and packages/cost/models/authors/xai/endpoints.ts:142-184

The model supports audio input but pricing structure doesn't account for it:

modality: { inputs: ["text", "image", "audio"], outputs: ["text"] }
// But pricing only has: input, output, web_search, no audio field

Recommendation: Either remove audio from modality or add audio pricing structure.

3. Endpoint Provider Model ID Mismatch - Low Priority

Location: packages/cost/models/authors/xai/endpoints.ts:143

Both reasoning and non-reasoning models use different provider model IDs:

• Reasoning: providerModelId: "grok-4-fast"
• Non-reasoning: providerModelId: "grok-4-fast-non-reasoning"

Verification needed: Confirm xAI's actual API model identifiers match these values.

4. Test Coverage Enhancement Opportunities - Low Priority

Location: worker/test/ai-gateway/registry-xai.spec.ts:284-400

Current tests are comprehensive but could add:

• Error handling for unsupported parameters (e.g., reasoning parameter on non-reasoning model)
• Validation of audio input handling
• Context length boundary testing (128K threshold)

Positive Aspects

✅ Excellent Test Coverage: Comprehensive test suite covering tool calls, response formats, and structured outputs
✅ Consistent Pricing Structure: Mirrors existing grok-4-fast-reasoning pricing patterns
✅ Proper Parameter Differentiation: Correctly excludes reasoning parameter from non-reasoning variant
✅ Snapshot Updates: Properly updated test snapshots reflect new model integration
✅ Security: No security concerns identified - standard model configuration patterns

Performance Considerations

• Context Length: 2M token context properly configured
• Pricing Thresholds: Appropriate 128K threshold for tiered pricing
• Token Limits: Reasonable maxCompletionTokens of 2M

Code Quality Assessment

• Type Safety: Proper TypeScript usage with satisfies constraints
• Documentation: Model descriptions are complete and informative
• Patterns: Follows established Helicone model integration patterns
• Maintainability: Clear separation of concerns between models and endpoints

Suggestions Summary

1. Fix model name formatting for consistency with other xAI models
2. Clarify audio modality support - either implement pricing or remove from inputs
3. Verify provider model IDs with xAI API documentation
4. Consider adding error case tests for parameter validation

Breaking Changes Assessment

No breaking changes identified - purely additive functionality.

[greptile-apps]

Greptile Overview

Summary

This PR integrates xAI's new `grok-4-fast-non-reasoning` model variant into the Helicone AI gateway, expanding the xAI provider's model portfolio. The changes add comprehensive support for the new model across three key areas of the codebase:

1. Model Definition: The new model is defined in packages/cost/models/authors/xai/models.ts with multimodal capabilities (text, image, and audio inputs), a 2M token context length, and appropriate metadata. The model is positioned as a faster, non-reasoning alternative to the existing grok-4-fast-reasoning model.

2. Cost Configuration: Endpoint configuration is added in packages/cost/models/authors/xai/endpoints.ts with tiered pricing based on context length (128K threshold), mirroring the pricing structure of the reasoning variant but excluding the reasoning parameter from supported parameters.

3. Testing Coverage: Comprehensive test suite added in worker/test/ai-gateway/registry-xai.spec.ts covering basic functionality, tool calls, response formats, structured outputs, and various error scenarios to ensure robust integration.

Additionally, the PR includes updates to the existing grok-4-fast-reasoning model: increasing maxCompletionTokens from 30,000 to 2,000,000 tokens to align with the model's actual capabilities, adding the reasoning parameter to distinguish it from the non-reasoning variant, and formatting consistency improvements by removing underscores from token numbers.

This integration follows Helicone's established patterns for adding new models and maintains compatibility with existing AI gateway functionality while providing users with more xAI model options tailored to different use cases.

PR Description Notes:

• Minor typo: "Tests & snapshots" should be "Tests and snapshots" for consistency

Important Files Changed

Changed Files

Filename	Score	Overview
worker/test/ai-gateway/registry-xai.spec.ts	5/5	Added comprehensive test suite for grok-4-fast-non-reasoning covering all features and error scenarios
packages/cost/models/authors/xai/endpoints.ts	4/5	Added endpoint configuration for new model and updated existing model parameters
packages/cost/models/authors/xai/models.ts	2/5	Added model definition but contains truncated description ending mid-sentence

Confidence score: 3/5

• This PR requires careful review due to content quality issues in model definitions
• Score lowered due to truncated model description that appears incomplete or corrupted
• Pay close attention to packages/cost/models/authors/xai/models.ts for the incomplete description field

_{3 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}