A4F
v20.0.0•

Provider 2 Major Model Surge, Vision Capabilities Restoration & Token Accuracy Enhancement

Sreejan
SreejanAuthor

This major update introduces 42 new chat/completion models to Provider 2 through a new API integration, significantly expanding model diversity across Qwen, DeepSeek, Hermes, GLM, Mistral, Gemma, and other families. A critical vision bug fix restores full multimodal functionality to 11 vision-capable models. Provider 7 receives accurate token counting using official provider data, and Provider 8 receives a major context window expansion.

Provider 2

43

New API Integration & Model Expansion (42 models)

Major Infrastructure Enhancement: New API integration enables access to 42 new chat/completion models spanning diverse model families with function calling, reasoning, and vision capabilities.

Qwen Models (13 models)

Free Tier:

Basic Tier:

Pro Tier:

DeepSeek Models (6 models)

Free Tier:

Basic Tier:

NousResearch Hermes Models (3 models)

Free Tier:

Pro Tier:

Zhipu AI GLM Models (5 models)

Free Tier:

Basic Tier:

OpenAI OSS Models (2 models)

Free Tier:

Moonshot AI Kimi Models (2 models)

Basic Tier:

Mistral Models (3 models)

Free Tier:

Basic Tier:

Google Gemma Models (3 models)

Free Tier:

Other Models (5 models)

Free Tier:

Basic Tier:

Vision Bug Fix (1 fix)

Critical Fix: Resolved a bug where multimodal/vision content was being stripped from API requests, rendering vision-capable models non-functional for image analysis.

11 Vision Models Restored:

Fix Details:

  • Multimodal content arrays (images, audio) now properly preserved
  • Base64 and URL image formats fully supported
  • Backward compatible with text-only requests

Provider 7

1

Token Counting Enhancement

Accuracy Improvement: Token counting now uses official provider data instead of estimation, with automatic fallback when needed.

  • Primary Source: Official token counts from provider API responses
  • Fallback Mechanism: Automatic fallback to estimation when provider data unavailable
  • Scope: Provider 7 only - no impact on other providers
  • Benefit: More accurate billing and usage tracking

Provider 8

1

Context Window Expansion

Performance Enhancement: Major context window expansion for improved long-document processing capabilities.

  • Context Window: Expanded from 40K to 262K tokens (~6x increase)
  • Tier Allocation: 131K (Free) / 183K (Basic) / 262K (Pro/Ultra)
  • Standard Rule Applied: 50%/70%/100%/100% distribution for free/basic/pro/ultra tiers

Feature Distribution Summary

3

New Model Capabilities Breakdown:

  • Function Calling: 35 models with tool/function calling support
  • Reasoning: 26 models with enhanced reasoning capabilities
  • Vision: 11 models with image analysis support

Platform Impact

5
  • Model Diversity: 42 new models across 9 model families provide unprecedented choice for specialized tasks
  • Vision Restoration: 11 vision-capable models now fully functional for image analysis workflows
  • Token Accuracy: Provider 7 users benefit from precise token counting for accurate usage tracking
  • Context Expansion: Provider 8 context window expansion enables processing of larger documents
  • Free Tier Value: 24 of the 42 new models available on Free tier, expanding accessible AI capabilities

Important Notes

4
  • Context Window Allocation: All new models follow the standard 50%/70%/100%/100% distribution rule for free/basic/pro/ultra tiers.
  • Vision Models: The vision bug fix applies only to Provider 2 models. Other providers' vision capabilities remain unchanged.
  • Token Counting: The token counting enhancement applies exclusively to Provider 7. Other providers continue using their existing token counting methods.
  • Model Availability: New models are available immediately. Check the Models page for complete specifications and tier availability.