Native Function Calling Now Live - Full Agentic Coding Support

Agentic Coding Support: Full native function/tool calling implementation enabling seamless integration with popular coding assistants.

Streaming with Tools Enabled

• Previously Blocked: Streaming requests with tools (function calling) were not supported
• Now Supported: Requests with stream: true and tools parameter work seamlessly
• Real-time Responses: Get real-time streaming of tool call results

Multi-Turn Function Calling Fixed

• Issue Resolved: Fixed compatibility issues with multi-turn function calling conversations
• Universal Compatibility: Works with Claude Code, Roo, Cline, GPT Codex, etc.
• OpenAI-Compatible: Functions anywhere OpenAI-compatible APIs are accepted

Provider 7 - Anthropic Tool Calling Integration

• Full Conversion: Complete tool message format conversion for Anthropic models
• Stream Support: Proper handling of tool_use content blocks in streaming responses
• Index Mapping Fix: Corrected index mapping for streaming tool call chunks

New Models Added (5 models)

Chat/Completion Models:

• provider-3/glm-4.7 - Pro Tier

  • Latest Z.AI model with 200K context, 128K output
  • Features: reasoning, function_calling

• provider-3/gemini-3-flash-preview - Pro Tier

  • Fast multimodal preview with 1M context
  • Features: reasoning, vision, function_calling, audio

• provider-3/minimax-m2 - Basic Tier

  • Compact MoE model for agentic workflows
  • Features: function_calling

Image Editing Models:

• provider-3/gemini-2.5-flash-image-preview-edit - Pro Tier

  • Natural language image editing

• provider-3/gpt-image-1-edit - Pro Tier

  • Targeted image modifications

Feature Updates (25+ models)

Function Calling Added:

• provider-3/kimi-k2, provider-3/kimi-k2-thinking, provider-3/kimi-k2-0905
• provider-3/qwen-3-max
• provider-3/gpt-4o-audio-preview

Audio Capability Added:

• provider-3/gemini-2.0-flash, provider-3/gemini-2.0-pro, provider-3/gemini-2.5-flash

Vision Capability Added:

• provider-3/gemma-3-27b-it, provider-3/gemma-3n-e4b-it, provider-3/gemma-3-4b-it, provider-3/gemma-3-12b-it, provider-3/gemma-3-1b-it
• provider-3/gpt-oss-120b, provider-3/gpt-oss-20b

Feature Corrections:

• provider-3/deepseek-r1-0528 - Removed reasoning
• provider-3/deepseek-v3 - Removed function_calling
• provider-3/deepseek-v3.1 - Added reasoning
• provider-3/glm-4.5, provider-3/glm-4.5-basic, provider-3/glm-4.6 - Feature corrections
• provider-3/sonar-reasoning, provider-3/sonar-reasoning-pro - Removed reasoning
• provider-3/qwen-3-coder-480b, provider-3/qwen3-235b, provider-3/llama-3.3-70b - Feature corrections

Tool Calling Compatibility Fix

• Fixed handling when assistant message content is null/missing with tool_calls
• Ensures content field exists for all tool role messages
• Resolves compatibility issues with models returning null content

Infrastructure Enhancement

• Load Balancing: Secondary API key added for improved reliability
• Rate Limit Mitigation: Automatic key rotation when limits are reached
• Higher Throughput: Effectively doubles available request capacity

Function Calling Rollout (11 models)

• provider-5/meta-llama-3.1-8b-instruct-fast
• provider-5/meta-llama-3.1-8b-instruct
• provider-5/kimi-k2-instruct
• provider-5/kimi-k2-thinking
• provider-5/qwen3-30b-a3b-thinking-2507
• provider-5/qwen3-coder-30b-a3b-instruct
• provider-5/qwen3-coder-480b-a35b-instruct
• provider-5/hermes-4-70b
• provider-5/qwen3-next-80b-a3b-thinking
• provider-5/intellect-3

Context Window Expansions

• provider-5/llama-guard-3-8b - 8x context increase
• provider-5/gemma-3-27b-it, provider-5/gemma-3-27b-it-fast - 13x context increase
• provider-5/deepseek-v3-0324 - 2.5x context increase
• provider-5/kimi-k2-thinking - 4x context increase
• provider-5/kimi-k2-instruct - Major expansion

Feature Additions

Reasoning Added:

• provider-5/qwen3-32b, provider-5/qwen3-32b-fast, provider-5/hermes-4-70b

Vision Added:

• provider-5/nemotron-nano-v2-12b
• provider-5/gemma-3-27b-it, provider-5/gemma-3-27b-it-fast

Embedding Models Improved (4 models)

• provider-5/bge-en-icl - 4x max tokens increase
• provider-5/bge-multilingual-gemma2 - Improved performance
• provider-5/e5-mistral-7b-instruct - Improved performance
• provider-5/qwen3-embedding-8b - 25% max tokens increase

Image Generation Optimized (2 models)

• provider-5/flux-dev - Improved efficiency
• provider-5/flux-schnell - Improved efficiency

HTTP Error Handling Improvements

Improved error handling across 7 providers for better debugging and consistent error reporting:

• Provider 1 - Added early HTTP error detection
• Provider 2 - Added early error detection with body reading
• Provider 3 - Added early HTTP error detection
• Provider 5 - Error body read before context closes
• Provider 6 - Error body read before context closes
• Provider 7 - Full response captured on error
• Provider 8 - Error body read before context closes

Improvements:

• Error responses now captured and logged correctly
• Consistent error handling pattern across all providers
• Error details preserved even for streaming requests

• Native Function Calling: Full agentic coding support across Claude Code, Roo, Cline, GPT Codex, etc.
• Streaming + Tools: Real-time function calling now fully supported
• Provider 7 Anthropic: Complete tool calling protocol conversion for Claude models
• Provider 3 Expansion: 5 new models + 25+ feature updates
• Provider 5 Load Balancing: Improved reliability and throughput

• Rolling Deployment: Function calling is progressively rolling out. Check the Models page to verify model support.
• Agentic Coding: Works with Claude Code, Roo, Cline, GPT Codex, and any OpenAI-compatible platform.
• Feature Flag Updates: Model capabilities have been corrected across Provider 3 and Provider 5.
• Context Windows: Several Provider 5 models received major context window expansions. Pro and Ultra tiers receive 100% allocation.