AI Provider and Model Updates: Adding New, Removing Deprecated, and Enhancing Functionality

[kqlio67]

# Removed from the framework or added to G4F

• Removed providers:

  • g4f/Provider/BlackboxAPI.py

• New providers added:

  • g4f/Provider/AllenAI.py (Close New Provider for Llama Tülü #2732)

• List of issues that can be closed:

  • Close Providers and models in version 0.4.7.0 #2735

---

# Docs (docs/)

docs(docs/providers-and-models.md): Update provider listings and model information

• Remove deprecated BlackboxAPI provider from no-auth section
• Add new AI models (grok-3, grok-3-r1) to Liaobots provider
• Include deepseek-v3 model in OIVSCode text models
• Update Gemini model references to version 2.0 specifications
• Adjust provider counts for multiple text models (gpt-4, deepseek-chat)
• Add new mixtral-small-24b and r1-1776 model entries
• Reorganize model order in text/image sections for consistency

chore(docs/providers-and-models.md): Remove deprecated Grok models from provider listings

• Remove Grok-2 model from Liaobots provider offerings
• Update Text Models section by eliminating discontinued Grok-2 entries
• Clean up duplicate Gemini model references in provider specifications

feat(docs/providers-and-models.md): update providers and models documentation

• Add new AllenAI provider to no-auth required section
• Introduce 5 new Ai2 text models in text models table
• Update DDG provider's text model from mixtral-small-28b to 24b variant

---

G4F (g4f/)

feat(g4f/models.py): update model configurations and expand provider support

• Add Grok-3 and Grok-3-R1 models under x.ai provider
• Introduce r1-1776 model for Perplexity AI platform
• Remove deprecated o1-mini model configuration
• Update provider lists for GPT-4 (remove Mhystical/Blackbox), Gemini 1.5 Pro (remove Blackbox), and DeepSeek-v3 (add OIVSCode)
• Optimize provider priorities by adjusting IterListProvider orderings across multiple models
• Correct Hermes-2-DPO configuration to use single Blackbox provider
• Remove BlackboxAPI references from qwq-32b and other affected models

chore(g4f/models.py): Remove Grok-2 and update Gemini provider configurations

• Remove Grok-2 model definition and all related references from x.ai section
• Exclude Liaobots from Gemini 1.5 Flash and Pro provider lists
• Update provider configurations for improved service accuracy
• Maintain consistent provider selection across related AI models

feat(g4f/models.py): Add Ai2 models and update provider references

• Add AllenAI provider import for no-auth required services
• Implement new Ai2 text models: tulu-3-405b, olmo-2-13b, tulu-3-1-8b, tulu-3-70b, and olmoe-0125
• Update 'evil' model provider from PollinationsAI to AllenAI
• Integrate new Ai2 models into ModelUtils conversion mapping

fix(g4f/models.py): update experimental model provider configuration

• Change evil model's best_provider from AllenAI to PollinationsAI
• Maintain experimental designation for uncensored AI model integration

---

# Provider (g4f/Provider/)

feat(g4f/Provider/Blackbox.py): update model configurations and premium handling

• Standardize model naming conventions (BLACKBOXAI instead of blackboxai)
• Update premium models list and restructure userSelectedModel composition
• Remove hardcoded premium user session data from request payload
• Add new 'o1' model to trendingAgentMode configurations
• Simplify message creation by removing timestamp injection for assistant messages
• Introduce 'beastMode' flag in API request parameters
• Clean up deprecated agent entries from trendingAgentMode

---

feat(g4f/Provider/ChatGLM.py): add finish reason handling and update default model

• Update default model from "all-tools-230b" to "glm-4"
• Add FinishReason import and yield stop reason when status is 'finish'
• Remove deprecated model_aliases configuration
• Refactor text content extraction for better readability
• Breaking: Remove legacy model alias "glm-4" mapping (now default)

---

chore(g4f/Provider/DDG.py): Reorder model entries for consistency

• Standardize ordering of models in DDG.models list
• Align model_aliases dictionary order with primary models list
  – Maintain identical functionality while improving code organization

---

feat(g4f/Provider/ImageLabs.py): Update default image model to sdxl-turbo

• Change default_model value from 'general' to 'sdxl-turbo'
• Remove redundant model_aliases mapping
• Update image_models and models attributes through new default value

---

feat(g4f/Provider/Liaobots.py): update supported model configurations and aliases

• Add new Grok-3 and Grok-3-Thinking models with enhanced token limits
• Update DeepSeek model keys to lowercase format for consistency
• Remove deprecated Gemini 1.5 models and add new Gemini 2.0 variants
• Add new o3-mini model configuration for OpenAI
• Simplify DeepSeek model aliases to use direct mappings
• Update Gemini alias mappings for flash-thinking and pro variants

fix(g4f/Provider/Liaobots.py): remove deprecated Gemini model aliases

• Remove "gemini-1.5-flash" and "gemini-1.5-pro" from model aliases
• Simplify Gemini alias mappings by eliminating obsolete version references
• Maintain consistent model alias structure across different providers

---

feat(g4f/Provider/OIVSCode.py): Update API endpoint and expand model support

• Update api_base URL to new Render subdomain
• Add DeepSeek-V3 to supported models list
• Include model alias mapping for DeepSeek-V3

---

fix(g4f/Provider/PollinationsAI.py): Resolve model initialization and alias conflicts

• Initialize text_models and image_models with default model values
• Remove duplicate model alias entries conflicting with default mappings
• Ensure fallback models exist when external model fetching fails

refactor(g4f/Provider/PollinationsAI.py): improve model initialization and error handling

• Add _models_loaded flag to prevent redundant API calls
• Restructure model combination logic to preserve default models
• Ensure default models remain available when API calls fail
• Call get_models() at start of create_async_generator
• Simplify model list deduplication using dict.fromkeys
• Add fallback to default models during exception handling

---

fix(g4f/Provider/PollinationsImage.py): Initialize image_models list and update label

• Initialize image_models class variable to empty list to prevent potential AttributeError
• Update class label from "Pollinations AI Image" to "Pollinations AI" for consistency
• Maintain existing model handling logic while ensuring proper list initialization in get_models

refactor(g4f/Provider/PollinationsImage.py): Improve model synchronization and initialization

• Initialize image_models with default model instead of empty list
• Add _models_loaded flag to track model synchronization status
• Combine parent class models with child class models in get_models
• Call get_models before async generation to ensure model availability

---

# Provider needs auth (g4f/Provider/needs_auth/)

fix(g4f/Provider/needs_auth/CablyAI.py): Enforce authentication requirement

• Change needs_auth class attribute from False to True to require API authentication

[github-actions]

Review for Pull Request: AI Provider and Model Updates

Summary

This pull request introduces significant updates to the AI providers and models within the G4F framework. It includes the addition of new models, removal of deprecated ones, and enhancements to existing functionalities across various components.

Changes Overview

1. Documentation Updates:

   • Removed deprecated BlackboxAPI provider.
   • Added new AI models (grok-3, grok-3-r1) to the Liaobots provider.
   • Updated references for the Gemini model to version 2.0 specifications.
   • Adjusted provider counts and reorganized model listings for consistency.

2. Model Configuration Updates:

   • Introduced new models under various providers, including Grok-3, r1-1776, and DeepSeek-V3.
   • Removed deprecated configurations and optimized provider priorities.

3. GUI Fixes:

   • Corrected the provider checkbox initialization logic to ensure proper evaluation of authentication conditions.

4. Provider Enhancements:

   • Standardized model naming conventions and improved premium handling.
   • Added new model configurations and simplified message creation processes.

5. Cache Validation Improvements:

   • Enhanced cache validation in the AsyncAuthedProvider to prevent writing invalid auth results.

Review Comments

• The removal of the BlackboxAPI provider is well-justified and helps streamline the codebase.
• The addition of new models and updates to existing ones reflect a thorough understanding of the current AI landscape.
• The changes to the GUI improve user experience by ensuring that the authentication requirements are clearly handled.
• The overall organization and clarity of the code have improved, making it easier to maintain and extend in the future.

Suggestions

• Consider adding unit tests for the new models and functionalities to ensure they work as expected and to prevent regressions in the future.
• It might be beneficial to include a changelog entry summarizing these updates for users who rely on the documentation.

Thank you for your contributions to the project! Your efforts in enhancing the G4F framework are greatly appreciated.

[github-actions]

The removal of o1_mini from the model mappings could potentially break integrations or scripts relying on its configuration. Verify that its deprecation has been communicated and managed effectively.

[github-actions]

The replacement of grok_2 with grok_3 in the model mappings is a breaking change. Ensure migration paths or backward compatibility considerations are adequately addressed.

[github-actions]

Use isinstance(auth_result, AuthResult) to check the type of auth_result before proceeding. Ensure that the logic for handling None values is clear and consistent.

[github-actions]

The condition elif auth_result is not None: seems redundant after the previous check. Consider simplifying the logic to avoid unnecessary checks.

[github-actions]

The unlink() method is called without checking if the file exists after the previous conditions. Ensure that the file exists before attempting to unlink it to avoid potential errors.

[github-actions]

Pull Request Review for: AI Provider and Model Updates: Adding New, Removing Deprecated, and Enhancing Functionality

Summary

This pull request contains significant updates, additions, and enhancements across many sections of the G4F framework. It aims to improve AI provider functionality, update model configurations, remove deprecated features, correct documentation, and improve code consistency. Major updates include the removal of BlackboxAPI, the introduction of AllenAI, and support for new models like Grok-3, Mixtral-small-24B, and DeepSeek-v3, among others. The changes also enhance documentation clarity by eliminating deprecated references and reordering content for better readability.

---

Key Updates

Providers

• Removed Providers:

  • g4f/Provider/BlackboxAPI.py removed as it was deprecated and no longer actively supported.

• Added Providers:

  • AllenAI.py introduces support for various new AI2 models.
  • Enhanced handling of existing AI providers, including Liaobots, ChatGLM, and OIVSCode.

Models

• Multiple new models supported under relevant providers:
  • New models introduced:
    • tulu-3-405b, tulu-3-70b, olmo-2-13b for Ai2.
    • Grok-3, Grok-3-r1 under x.ai.
    • DeepSeek-v3 model in OIVSCode and DDG updated to mixtral-small-24b.
    • Updated Gemini models to Gemini-2.0.
  • Deprecated models, such as Grok-2 and outdated Gemini-1.5 Flash/Pro, were removed from the offerings.

Documentation

• Enhanced Documentation:
  • Removed references to BlackboxAPI and Grok-2 from the model listing.
  • Adjusted provider counts, descriptions, and tables to remove duplicates and include new entries like AllenAI.
  • Reorganized model tables (text and image models) for better consistency.

Configuration

• Expanded g4f/models.py model configurations and ensured consistency in providers mapping, like correcting evil model's provider.
• Improved provider model alias mappings, with clear deprecation of outdated mappings.
• Added consistent error handling and better metrics tracking in experimental providers like AllenAI.

Bug Fixes

• Corrected logical issues with the initialization of provider checkboxes in g4f/gui/client/static/js/chat.v1.js.
• Improved cache validation in g4f/providers/base_provider.py to prevent writing invalid auth results to cache.

---

Strengths of the PR

1. Support for Cutting-Edge Models: By introducing support for newly released models and providers like AllenAI, the framework stays up-to-date with the latest AI advancements.
2. Code Consistency: Deprecated code and models were removed, which simplifies the framework and reduces technical debt.
3. Enhanced Clarity: Documentation updates ensure users and contributors are not confused by redundant or outdated information.
4. Bug Fixes and Stability: Meaningful updates to front-end logic, critical API configurations, and caching mechanisms increase reliability across the framework.

---

Suggestions for Improvement

Below are some minor areas for improvement or clarification:

1. Testing:

   • Verify that the newly added providers (particularly AllenAI and its models) are producing results as expected across different use cases.
   • Test breaking changes such as ChatGLM's default model migration and CablyAI's enforced auth.

2. Docs:

   • Provide a migration note or changelog entry for BlackboxAPI users to help them transition to an alternative provider.
   • Clarify "breaking change" notices in the PR description so teams can better adapt their usage, especially when associated models like Grok-2 and Gemini are deprecated.

3. Code Comments:

   • While most sections are self-explanatory, newly added providers like AllenAI include complex structures for request/response generation. Adding inline explanations would help future contributors understand the design.

---

Merge Approval

This PR introduces significant and useful changes that improve the framework's usability, capability, and reliability. Once the above testing and documentation suggestions are addressed, this PR is ready to merge.

---

Reviewer Notes

Thank you for your substantial contribution to the G4F project! These updates will greatly extend its capabilities and maintain its relevance with current AI provider technologies.

---

Reviewed by: G4F Copilot

[github-actions]

Review for Pull Request #2735

Title: AI Provider and Model Updates: Adding New, Removing Deprecated, and Enhancing Functionality

---

Thank you for your detailed contribution to this project. This pull request includes significant changes to the framework, including the removal of deprecated providers, the addition of new providers and models, and various improvements to documentation and functionality. Below is my comprehensive review of the proposed changes.

---

Summary of Changes

1. Removed Providers and Models

• The BlackboxAPI provider has been removed due to its deprecation.
• Deprecated models such as o1-mini and Grok-2 have been cleansed from the framework.

2. Added Providers and Models

• Introduced the new provider AllenAI in g4f/Provider/AllenAI.py, offering no-auth-required services and support for Ai2 models like tulu-3-405b and olmo-2-13b.
• Newly added models include Grok-3, Grok-3-R1, mixtral-small-24b, and r1-1776 with appropriate provider assignments.

3. Documentation Enhancements

• Updated the docs/providers-and-models.md file to reflect the newly added and removed providers/models, along with reorganizing model entries for improved consistency.

4. Bug Fixes and Code Improvements

• Fixed the logic for provider checkboxes in g4f/gui/client/static/js/chat.v1.js to properly handle auth-required providers.
• Improved cache validation in AsyncAuthedProvider by adding type checks to prevent runtime errors.
• Enhanced model initialization logic in PollinationsAI.py and PollinationsImage.py to handle API failures gracefully while maintaining fallback defaults.

5. Breaking Changes

• Changed the default model in ChatGLM from all-tools-230b to glm-4, while removing support for legacy aliases.
• Removed hardcoded premium user simulation in the Blackbox provider.

---

Detailed Feedback and Suggestions

1. Code Quality:

   • The restructuring of provider configurations (e.g., Blackbox.py, Liaobots.py) is well-organized.
   • Most of the changes adhere to best practices, especially the improvements in error handling (e.g., _models_loaded flag usage in PollinationsAI.py).

2. Documentation:

   • The documentation updates in providers-and-models.md are clear and comprehensive.
   • Small phrasing improvements may help clarify the purpose of some new models or provider entries—for example, more detail on model token limits in the text models table.

3. Backward Compatibility:

   • While the removal of BlackboxAPI and older models such as o1-mini is justified, consider if they might require deprecation warnings in previous releases for smoother transitioning.

4. Testing Consideration:

   • Ensure that the newly added providers and models (e.g., AllenAI, tulu-3-405b) have been tested for integration with core functionality and edge cases.
   • Verify that the new checkbox logic in chat.v1.js works correctly for all auth configurations.

5. Optimization Opportunities:

   • In PollinationsAI.py, reduce duplicate calls to get_models() by caching results when _models_loaded is set.
   • Consider consolidating duplicate structures or constants in Liaobots.py to improve maintainability.

---

Actionable Changes

• Documentation: Expand on the reasoning behind breaking changes in the pull request description.
• Backward Compatibility: Add deprecation reminders, logs, or transitional notes where applicable.
• Testing: Confirm all supported models and providers (including experimental changes like evil model updates) are validated in CI/CD workflows before merging.

---

Conclusion

Overall, this pull request significantly enhances the capabilities and maintainability of the g4f framework. Although a few minor improvements can be made, it is a well-executed update. Once testing confirms functionality, these changes will add considerable value to the project.

Thank you again for your thoughtful contributions. Please feel free to address any points above before merging the pull request.

— g4f Copilot

[github-actions]

Review by g4f Copilot

Thank you for your extensive contribution to the G4F repository! This pull request introduces several provider and model updates, documentation improvements, and fixes that enhance the project's functionality and maintainability. Below are the detailed observations and feedback regarding the changes:

---

Summary of Changes Reviewed

1. Removed Providers

• The removal of BlackboxAPI.py is clear and justified based on its deprecated status.

2. Added New Providers

• Introduction of AllenAI.py with its new models is a valuable addition. This aligns with modern requirements and improves overall feature diversity.

3. Documentation Updates

• Clarity in Updates
  • The updates to providers-and-models.md correctly reflect the changes made to the models and providers. The restructuring and consistency improvements make the documentation more understandable.
  • The removal of deprecated entries (e.g., Grok-2) and inclusion of new models such as AllenAI’s models (tulu-3-405b, olmo-2-13b, r1-1776, etc.) improves accuracy.
• Breaking Changes Highlighted
  • Breaking changes (e.g., removal of legacy aliases) are well-documented. However, it may help to provide a specific section in the documentation summarizing all breaking changes to aid user migration.

4. Codebase Enhancements

Provider Improvements

• Blackbox.py
  • Renaming to uppercase BLACKBOXAI provides better readability and standardization.
  • Removal of hardcoded premium session data and implementation of a beastMode flag are significant improvements that enhance security and reduce unwanted dependencies.
• AllenAI.py
  • The new provider is well-structured and follows a consistent coding style. The use of a custom Conversation class coupled with support for optional parameters increases flexibility. The multipart form implementation is also clear.
• ChatGLM.py
  • Addition of FinishReason handling helps streamline the process of detecting the end of a conversation. This enhances the robustness of the provider.
• PollinationsAI.py and PollinationsImage.py
  • Synchronization between parent and child models is handled effectively with the _models_loaded flag.
  • Enhanced fallback logic during API failures ensures improved reliability for users.
• Liaobots.py
  • Support for new Grok-3 models, removal of deprecated models (e.g., Gemini 1.5), and improved alias handling increase compatibility and usability.
• Other Providers
  • Updates to model aliases, API endpoints (e.g., OIVSCode.py), and consistency changes are appropriate.

Auth & Cache Improvements

• Handling of invalid AuthResult types in AsyncAuthedProvider represents a solid step in ensuring accurate cache storage. Validation checks before writing to disk prevent potential bugs.

GUI Fix

• Transitioning the ternary operator in chat.v1.js to a clearer logic for evaluating checked attributes aligns with better coding practices and prevents initialization errors.

5. Model Updates

• Updates to g4f/models.py improve model configurations, support new AllenAI, Grok-3, and r1-1776 models while removing deprecated models (o1-mini, Grok-2). Optimizing priority order with IterListProvider ensures users benefit from the best-performing providers.

---

Overall Feedback

✅ Strengths

1. Well-Structured Additions: New features fit well into the existing framework and provide value without disrupting current functionality.
2. Code Quality: The changes maintain consistency with the existing style, avoiding unnecessary complexity.
3. Documentation: The updates provide accurate and detailed information on the changes, ensuring users can easily understand the new system.
4. Backward Compatibility Awareness: Despite breaking changes, efforts have been made to minimize user impact. Clear documentation mitigates migration challenges.
5. Focus on Reliability & Security: Removing hardcoded session data and improving fallback mechanisms ensure a more robust experience.

⚠️ Suggestions for Improvement

1. Breaking Changes Summary
   • Although individual breaking changes are mentioned in the documentation, presenting them as a consolidated list would improve clarity.
2. Test Coverage
   • Including (or documenting) test results for the new providers, models, and updated configurations can assure contributors of the stability of these changes.
3. Readability in Form Data Creation
   • In AllenAI.py, the multipart form data creation can be simplified with a utility function to improve readability and reduce boilerplate.
4. Explicit Class-Level Comments
   • While methods are well-documented, class-level docstrings (especially for Conversation and new provider classes) would make the codebase more intuitive for contributors.

---

Conclusion

This is a well-executed pull request that enhances G4F's functionality and keeps the project up-to-date with modern AI providers and models. After addressing the minor suggestions above (testing discussion and breaking change summary), it is ready to merge.

Thank you again for your contribution to advancing the project! 🎉

Best regards,
g4f Copilot

[TheFirstNoob]

@kqlio67 Hey, i think need disable Prodia or update to api v2 with request key (need_auth). API v1 disabled and all models not avalable

[github-actions]

Pull Request Review: AI Provider and Model Updates

Overview

Thank you for contributing to the project! This pull request introduces significant changes and improvements, including the removal of deprecated elements, the addition of new providers and models, and documentation updates to enhance usability and functionality. Below, I've outlined the key points of the review based on the changes made:

---

Strengths

1. Provider Enhancements:

   • Great to see outdated providers, such as BlackboxAPI.py, being removed for a cleaner codebase.
   • The addition of new providers (AllenAI, for example) and models (including Ai2 models such as tulu-3-405b and olmo-2-13b) is well-documented and helps expand the framework's offerings.

2. Documentation Updates:

   • Changes to providers-and-models.md provide critical updates to users, detailing deprecated providers and newly introduced models. This ensures users remain well-informed.

3. Model Optimizations:

   • Aligning with updated provider offerings like the gemini-2.0 integration and reconfiguring priorities for models like Grok-3 ensures logical improvements in service accuracy and retrieval efficiency.

4. Bug Fixes:

   • Resolved ternary operator issues in g4f/gui/client/static/js/chat.v1.js.
   • Addressed issues in provider initialization to ensure robust fallbacks in edge cases (PollinationsAI and PollinationsImage).

5. Breaking Changes Awareness:

   • The ChatGLM.py breaking change ("glm-4" model mapping removed) is clearly documented, helping users adapt to the new system.

---

Changes That Can Be Improved

1. Consistency in Configuration Usage:

   • The removal of o1-mini might break expectations for users currently relying on this model. Consider adding a roadmap section in the documentation outlining deprecated models and alternatives.

2. Testing/Validation Plan:

   • The changes involve new models, providers, and API updates (e.g., AllenAI, DeepSeek-V3, etc.). It would be helpful to include a validation plan or information about tests conducted to verify these additions work as intended across success and error scenarios.

3. Supporting Model Aliases:

   • Some models (mixtral-small-24b, llama-3.3, etc.) have renamed aliases, but alias consistency checks were inconsistently handled in parts of the PR. A more robust alias mapping routine could be added to ensure backward compatibility and avoid redundant alias definitions.

4. Guard Clause for Optional Imports:

   • With the addition of several features (like AllenAI), ensuring fail-safe imports in environments missing their corresponding dependencies is essential for maintaining compatibility.

---

Recommendations Before Merging

• Documentation:

  • Ensure user-facing documentation explicitly lists all models/providers removed (o1-mini, Grok-2, etc.) and suggests alternatives.

• Regression Testing Results:

  • Validate how the removal of key provider imports (BlackboxAPI, for example) impacts downstream workflows that depend on them.

• API Endpoint Updates:

  • Ensure test environments for updated APIs (e.g., AllenAI's olmo-api.allen.ai/v4/message/stream) confirm reliability, especially under high network latency or edge cases.

---

Conclusion

This pull request successfully enhances the framework while removing deprecated providers, improving configurations, and expanding documentation. Addressing the changes recommended above will ensure a smooth user experience while maintaining backward compatibility.

Thank you for this comprehensive and thoughtful contribution!