# Model Deprecations

This page tracks deprecated models and upcoming deprecations. We recommend migrating to newer, more capable models listed in our recommendations.

Deprecated models are no longer available through Pulze. Please migrate to recommended alternatives to ensure continued service.

## Migration Support

Need help migrating from deprecated models? Contact us at [support@pulze.ai](mailto:support@pulze.ai) to:

- Discuss your use cases and requirements
- Identify optimal replacement models
- Test new models before full migration
- Plan smooth transitions for your team

## Currently Deprecated Models

These models are no longer available:

### Anthropic

Claude 3.5 Sonnet (October 22, 2025)

**Deprecated**: October 22, 2025
**Recommended Alternatives**:

- **Claude Opus 4.1**: Very capable model with highest intelligence
- **Claude Sonnet 4.5**: Best for complex agents and coding tasks
- **Claude Sonnet 4.0**: Updated with 1M token context window (up from 200K)

**Migration Notes**: Claude 4 series provides significant improvements in reasoning, coding, and long-context handling.

### OpenAI

o1-preview

**Deprecated**: Recently
**Recommended Alternatives**:

- **GPT-5**: Flagship model with 400K context window
- **GPT-5 Mini**: Cost-efficient with high reasoning capabilities  
- **o3-deep-research**: For complex multi-step research tasks

**Migration Notes**: GPT-5 series offers superior performance with larger context windows and better reasoning.

o1-mini

**Deprecated**: Recently
**Recommended Alternatives**:

- **GPT-5 Mini**: Cost-efficient with high reasoning capabilities  
- **GPT-5 Nano**: Fastest, most affordable option

**Migration Notes**: GPT-5 Mini provides better performance at competitive pricing.

### Google

Gemini 1.5 Flash

**Deprecated**: Recently
**Recommended Alternatives**:

- **Gemini 2.5 Flash**: Best price/performance with thinking capabilities
- **Gemini 2.5 Pro**: Advanced reasoning with 1M+ context window

**Migration Notes**: Gemini 2.5 series provides significant improvements in reasoning and cost-efficiency.

Gemini 1.5 Pro

**Deprecated**: Recently
**Recommended Alternatives**:

- **Gemini 2.5 Pro**: Advanced reasoning with 1M+ context window
- **Gemini 2.5 Flash**: Best price/performance with thinking capabilities

**Migration Notes**: Gemini 2.5 Pro offers better reasoning and longer context windows.

Gemini 1.5 Experimental

**Deprecated**: Recently
**Recommended Alternatives**:

- **Gemini 2.5 Pro**: Advanced reasoning with 1M+ context window
- **Gemini 2.5 Flash**: Best price/performance with thinking capabilities

**Migration Notes**: Gemini 2.5 series represents the production-ready evolution of experimental features.

### Groq

DeepSeek R1 Distill

**Deprecated**: Recently
**Recommended Alternatives**:

- **Groq Compound**: Integrated tool orchestration
- **OpenAI GPT-OSS 120B**: Open-weight model with strong performance

**Migration Notes**: Groq Compound provides better tool integration capabilities.

Llama 3 Models

**Deprecated**: Recently
**Recommended Alternatives**:

- **Llama 4 Maverick**: Meta’s latest multimodal flagship
- **Llama 4 Scout**: Balanced performance and efficiency
- **OpenAI GPT-OSS Models**: 20B & 120B variants

**Migration Notes**: Llama 4 series provides significant improvements over Llama 3.

### Fireworks

Llama 3.2

**Deprecated**: Recently
**Recommended Alternatives**:

- **Llama 4 Models**: Via Groq (Maverick & Scout)
- **GPT-OSS Series**: Via Fireworks
- **DeepSeek V3.1 Variants**: Via Fireworks

**Migration Notes**: Llama 4 provides better performance across all benchmarks.

DeepSeek R1

**Deprecated**: Recently
**Recommended Alternatives**:

- **DeepSeek V3.1 Variants**: Available via Fireworks
- **GPT-OSS Series**: Available via Fireworks

**Migration Notes**: DeepSeek V3.1 offers improved reasoning and performance.

Phi 3.5

**Deprecated**: Recently
**Recommended Alternatives**:

- **GPT-5 Nano**: For fast, affordable inference
- **Gemini 2.5 Flash**: Best price/performance

**Migration Notes**: GPT-5 Nano provides better performance at similar cost points.

### Cohere

Command R

**Deprecated**: Recently
**Recommended Alternatives**:

- **Command A 03-2025**: 256K context, 111B parameters

**Migration Notes**: Command A provides significant improvements in reasoning and performance.

Command R+

**Deprecated**: Recently
**Recommended Alternatives**:

- **Command A 03-2025**: 256K context, 111B parameters

**Migration Notes**: Command A represents the next generation with better capabilities.

### Together AI

Various Legacy Models

**Deprecated**: Recently
**Recommended Alternatives**: Contact Pulze team for specific recommendations based on your use case.
**Migration Notes**: We can help identify optimal replacements from our current model catalog.

### Mistral

Large 2407

**Deprecated**: Recently
**Recommended Alternatives**: Contact Pulze team for recommendations based on your specific use case.
**Migration Notes**: Multiple strong alternatives available across providers.

Mixtral 8x22B

**Deprecated**: Recently
**Recommended Alternatives**: Contact Pulze team for recommendations based on your specific use case.
**Migration Notes**: Newer models offer better performance and capabilities.

## Upcoming Deprecations

These models will be deprecated on the specified dates:

Plan migrations before these deprecation dates to avoid service disruption.

### Anthropic

Claude 3 Opus - January 5, 2026

**Deprecation Date**: January 5, 2026
**Recommended Alternatives**:

- **Claude Opus 4.1**: Direct upgrade with significant improvements
- **Claude Sonnet 4.5**: Best for complex agents and coding tasks

**Action Required**: Begin testing Claude 4 series models and plan migration before January 5, 2026.

**Migration Impact**:
- Improved reasoning capabilities
- Better long-context handling
- Enhanced coding abilities
- More reliable tool use

### Google

Gemini 2.0 Flash Models - February 5, 2026

**Deprecation Date**: February 5, 2026
**Recommended Alternatives**:

- **Gemini 2.5 Flash**: Direct upgrade with better price/performance
- **Gemini 2.5 Pro**: For more complex reasoning tasks

**Action Required**: Begin testing Gemini 2.5 series and plan migration before February 5, 2026.

**Migration Impact**:
- Improved thinking capabilities
- Better reasoning performance
- Enhanced multimodal understanding
- Competitive pricing

## Recommended Models for Evaluation

When migrating or selecting new models, we recommend evaluating these current best-in-class options:

### Top Performers by Category

**Reasoning & General Intelligence**

- **xAI Grok 4 Fast**: Currently leading all benchmarks as best model in the world (2M context)
- **OpenAI GPT-5**: Flagship model with 400K context window
- **Anthropic Claude Opus 4.1**: Very capable with highest intelligence

**Coding & Agentic Tasks**

- **Anthropic Claude Sonnet 4.5**: Best for complex agents and coding
- **OpenAI GPT-5 Codex**: Optimized for agentic coding environments
- **xAI Grok Code Fast**: Advanced for agentic coding tasks

**Cost-Performance Leaders**

- **Google Gemini 2.5 Flash**: Best price/performance with thinking capabilities
- **OpenAI GPT-5 Nano**: Fastest, most affordable option
- **OpenAI GPT-5 Mini**: Cost-efficient with high reasoning capabilities

**Research & Analysis**

- **OpenAI o3-deep-research**: Complex multi-step research tasks
- **Google Gemini 2.5 Pro**: Advanced reasoning with 1M+ context window

**Specialized Use Cases**

- **AI21Labs Jamba Large 1.7**: Novel Mamba-Transformer architecture, 256K context
- **Groq Compound**: Integrated tool orchestration (web search, code execution, browser automation)

## Deprecation Policy

### How We Handle Deprecations

1. **Advance Notice**: Minimum 90 days notice before deprecation
2. **Migration Support**: Technical guidance and testing assistance
3. **Recommendations**: Clear alternatives for each deprecated model
4. **Gradual Transition**: Staged rollout of replacements

### Why Models Are Deprecated

Models may be deprecated due to:

- **Provider Updates**: Model provider releases significantly better versions
- **Performance**: Newer models substantially outperform on benchmarks
- **Cost**: Better price/performance alternatives become available
- **Capabilities**: New models offer features not in older versions
- **Provider Sunset**: Original provider discontinues the model

### Staying Informed

- **Email Notifications**: We email all affected organizations before deprecations
- **Documentation**: This page is kept current with all deprecation information
- **Support Channel**: Contact [support@pulze.ai](mailto:support@pulze.ai) with migration questions
- **Slack/Discord**: Join our community for real-time updates

## Migration Best Practices

### Before Migrating

1. **Review Use Cases**: Understand how you’re using the deprecated model
2. **Evaluate Alternatives**: Test recommended replacements with your workloads
3. **Check Compatibility**: Ensure new models support your required features
4. **Plan Timeline**: Allow sufficient time for testing and rollout

### During Migration

1. **Start with Test Spaces**: Validate in non-production environments first
2. **Run Evaluations**: Compare new model performance against your datasets
3. **Monitor Closely**: Track quality, cost, and performance metrics
4. **Gather Feedback**: Get user input on new model performance

### After Migration

1. **Monitor Performance**: Track ongoing quality and cost metrics
2. **Adjust Routing**: Update custom routers if applicable
3. **Document Changes**: Update internal documentation and playbooks
4. **Share Learnings**: Help your team understand any behavioral differences

## Need Help?

Our team is here to support your migration:

[**Contact Support**](mailto:support@pulze.ai)  
Email [support@pulze.ai](mailto:support@pulze.ai) for migration assistance.