Free Tier Expansion & Gemini 3 Pro Launch
This transformative update represents our largest free tier expansion to date, democratizing access to 24 cutting-edge AI models while introducing Google's groundbreaking Gemini 3 Pro Preview - the world's current SOTA model launched just 2 days ago. We've upgraded our provider infrastructure by replacing the legacy Provider 2 with a new, highly stable system and strategically promoted 9 flagship models to the free tier, ensuring everyone has access to state-of-the-art AI capabilities.
Gemini 3 Pro Preview - Ultra Tier Exclusive
World's Most Advanced AI Model: Google's groundbreaking Gemini 3 Pro Preview represents the current state-of-the-art in AI capabilities, launched just 2 days ago and demonstrating unprecedented performance across all major benchmarks.
- Ultra Tier Exclusive:
provider-2/gemini-3-pro-preview- SOTA flagship model- Benchmark Excellence: Industry-leading performance across reasoning, mathematics, coding, and multimodal tasks
- Cutting-Edge Capabilities: Next-generation architecture with enhanced reasoning depth and broader knowledge coverage
- Multimodal Mastery: Advanced vision, text, and audio understanding in a single unified model
- Production Ready: Despite "preview" status, offers production-grade reliability and consistent performance
- Exclusive Access: Available exclusively to Ultra tier subscribers as the platform's premier flagship offering
Provider 1 - Free Tier Expansion
Strategic Free Tier Promotion: 9 flagship models now accessible to all users, expanding free tier access to advanced reasoning, multimodal, and specialized capabilities.
9 Models Promoted to FREE Tier:
- Meta Llama Family:
provider-1/llama-4-maverick-17b-128e-instruct- Next-gen 17B Mixture-of-Experts flagship
- Enhanced instruction following and reasoning capabilities
provider-1/llama-3.3-swallow-70b-instruct-v0.4- Specialized 70B variant with enhanced multilingual support
- Optimized for complex reasoning tasks
- DeepSeek Advanced Reasoning:
provider-1/deepseek-r1-0528- R1 reasoning checkpoint with improved logic capabilities
- Chain-of-thought reasoning optimization
provider-1/deepseek-r1-distill-llama-70b- 70B distilled reasoning model
- Efficient inference with preserved reasoning quality
provider-1/deepseek-v3-0324- V3 0324 checkpoint with enhanced performance
- Improved general-purpose capabilities
- Qwen3 Series:
provider-1/qwen3-235b-a22b-instruct-2507- 235B flagship with advanced instruction following
- Multilingual excellence across 29 languages
- Mistral AI:
provider-1/mistral-small-3.1-24b-instruct-2503- 24B flagship with optimized efficiency
- High-quality generation at reduced computational cost
- Specialized Models:
provider-1/ax-4.0- SKT's AX 4.0 flagship model
- Korean language specialization with multilingual support
provider-1/hyperclovax-seed-think-14b- NAVER's HyperCLOVA X thinking-optimized model
- Advanced reasoning architecture with Korean expertise
Provider 2 - Infrastructure Replacement & Expansion
Major Infrastructure Update: Replaced legacy Provider 2 (bankrupt due to extreme API demand) with new, highly stable provider infrastructure featuring 17 advanced models.
15 New FREE Tier Models:
- Meta Llama Vision:
provider-2/llama-3.2-11b-vision-instruct- 11B vision-language model with multimodal capabilities
- Native image understanding and analysis
- Instruction-tuned for visual question answering
- Google Gemini:
provider-2/gemini-2.5-flash- Latest Flash variant with enhanced speed and efficiency
- Optimized for low-latency applications
- Multimodal text and image processing
- Microsoft Models:
provider-2/phi-3-medium-128k-instruct- 128K context window instruction model
- Efficient reasoning in compact architecture
provider-2/dialogpt-medium- Conversational AI medium variant
- Natural dialogue generation
provider-2/dialogpt-large- Large-scale dialogue generation
- Enhanced conversational coherence
- Mistral AI:
provider-2/codestral-2508- Code generation specialist
- Multi-language programming support
provider-2/mistral-medium-2505- Medium-scale flagship
- Balanced performance and efficiency
provider-2/mistral-nemo-instruct-2407- Instruction-tuned Nemo variant
- Compact yet powerful architecture
- DeepSeek Specialized:
provider-2/deepseek-r1-distill-qwen-1.5b- Compact 1.5B distilled reasoning model
- Efficient reasoning in minimal footprint
provider-2/deepseek-math-7b-instruct- Mathematical reasoning specialist
- Advanced problem-solving capabilities
- Google Gemma:
provider-2/gemma-3-4b-it- 4B instruction-tuned model
- Efficient general-purpose capabilities
- Qwen Coding & General:
provider-2/qwen2.5-coder-14b-instruct- 14B coding specialist
- Optimized for code generation and debugging
provider-2/qwen3-4b-instruct-2507- Compact 4B flagship
- Efficient instruction following
provider-2/qwen3-1.7b- Ultra-efficient 1.7B model
- Resource-friendly deployment
- NVIDIA:
provider-2/llama-3.1-nemotron-8b-ultralong-1m-instruct- 1M token context window model
- Ultra-long document processing
- Advanced memory optimization
2 Paid Tier Models:
- Pro Tier:
provider-2/gemini-2.5-pro- Professional-grade Gemini flagship
- Enhanced reasoning and generation quality
- Multimodal capabilities with vision support
- Ultra Tier Exclusive:
provider-2/gemini-3-pro-preview- World's current SOTA model launched 2 days ago
- Industry-leading benchmark performance across all domains
- Next-generation architecture with enhanced reasoning depth
- Advanced multimodal understanding (vision, text, audio)
- Production-grade reliability despite preview status
- Available exclusively on ULTRA tier
Legacy Provider 2 Models Removed
Complete Provider Retirement: All 104 models from bankrupt legacy Provider 2 have been discontinued due to provider bankruptcy and depleted credits from extreme API demand. The new Provider 2 infrastructure offers superior stability with carefully curated model selection.
104 Models Removed:
- DeepSeek Models (17 models):
provider-2/deepseek-chat,provider-2/deepseek-r1,provider-2/deepseek-r1-0528,provider-2/deepseek-r1-2025-01-20,provider-2/deepseek-r1-250120,provider-2/deepseek-r1-250528,provider-2/deepseek-r1-searching,provider-2/deepseek-reasoner,provider-2/deepseek-v3,provider-2/deepseek-v3-0324,provider-2/deepseek-v3-1,provider-2/deepseek-v3-1-250821,provider-2/deepseek-v3-1-think-250821,provider-2/deepseek-v3-250324,provider-2/deepseek-v3-search,provider-2/deepseek-v3.1,provider-2/deepseek-v3.2-exp
- Google Gemini Models (20 models):
provider-2/gemini-2.0-flash,provider-2/gemini-2.0-flash-001,provider-2/gemini-2.0-flash-lite-001,provider-2/gemini-2.5-flash,provider-2/gemini-2.5-flash-image,provider-2/gemini-2.5-flash-image-preview,provider-2/gemini-2.5-flash-lite,provider-2/gemini-2.5-flash-lite-nothinking,provider-2/gemini-2.5-flash-lite-preview-06-17,provider-2/gemini-2.5-flash-lite-preview-06-17-nothinking,provider-2/gemini-2.5-flash-lite-preview-06-17-thinking,provider-2/gemini-2.5-flash-lite-preview-09-2025,provider-2/gemini-2.5-flash-lite-thinking,provider-2/gemini-2.5-flash-nothinking,provider-2/gemini-2.5-flash-preview-09-2025,provider-2/gemini-2.5-flash-thinking,provider-2/gemini-flash-lite-latest
- Zhipu AI GLM Models (8 models):
- OpenAI GPT Models (14 models):
provider-2/gpt-4.1-mini,provider-2/gpt-4.1-mini-2025-04-14,provider-2/gpt-4.1-nano,provider-2/gpt-4.1-nano-2025-04-14,provider-2/gpt-4o-mini,provider-2/gpt-4o-mini-2024-07-18,provider-2/gpt-5-mini,provider-2/gpt-5-mini-2025-08-07,provider-2/gpt-5-nano,provider-2/gpt-5-nano-2025-08-07,provider-2/gpt-5.1-codex-mini,provider-2/gpt-5.1-codex-mini-2025-11-13,provider-2/gpt-oss-120b,provider-2/gpt-oss-20b
- Qwen Models (24 models):
provider-2/qvq-32b,provider-2/qvq-72b-preview,provider-2/qwen-max,provider-2/qwen-max-2025-01-25,provider-2/qwen-max-latest,provider-2/qwen-plus-latest,provider-2/qwen-vl-max,provider-2/qwen2.5-32b-instruct,provider-2/qwen2.5-72b-instruct,provider-2/qwen2.5-7b-instruct,provider-2/qwen2.5-vl-32b-instruct,provider-2/qwen2.5-vl-72b-instruct,provider-2/qwen3-235b-a22b,provider-2/qwen3-235b-a22b-think,provider-2/qwen3-30b-a3b,provider-2/qwen3-30b-a3b-think,provider-2/qwen3-32b,provider-2/qwen3-coder-plus,provider-2/qwen3-max,provider-2/qwen3-max-preview-n,provider-2/qwen3-next-80b-a3b-instruct,provider-2/qwen3-vl-235b-a22b-instruct,provider-2/qwen3-vl-plus,provider-2/qwq-plus
- Meta Llama Models (6 models):
- Baidu ERNIE Models (6 models):
- Moonshot AI Kimi Models (4 models):
- Perplexity Models (2 models):
- Other Models (5 models):
provider-2/doubao-seedream-4-0-250828(Doubao)provider-2/hunyuan-t1(Tencent)provider-2/MAI-DS-R1(MAI)provider-2/mistral-small-latest(Mistral)provider-2/SparkDesk-v3.5(iFlytek)
Alternative Access: Most removed models remain available through Provider 1, Provider 3, and other providers with identical or enhanced capabilities and improved stability.
Platform Impact & Benefits
- Free Tier Democratization: 24 models now accessible to all users, the largest free tier expansion in platform history
- Enhanced Reliability: New Provider 2 infrastructure offers improved uptime, reduced latency, and consistent performance
- Model Diversity: Comprehensive coverage across reasoning (DeepSeek R1), vision (Llama 3.2 Vision), coding (Codestral, Qwen Coder), mathematics (DeepSeek Math), and general chat capabilities
- Zero Migration Required: All tier promotions are automatic upgrades—existing free tier users gain immediate access to 9 additional premium models
Technical Highlights
- Reasoning Excellence: DeepSeek R1 and HyperCLOVA X models bring advanced chain-of-thought capabilities to free tier
- Vision & Multimodal: Llama 3.2 11B Vision enables image understanding and analysis for all users
- Extended Context: NVIDIA Nemotron with 1M token context window for ultra-long document processing
- Coding Specialists: Codestral and Qwen Coder models optimized for code generation and completion
- Mathematical Reasoning: DeepSeek Math 7B specialized for advanced mathematical problem-solving
Model Distribution Summary
- Total Models Added: 26 new models across both providers
- Free Tier Access: 24 models (9 from Provider 1 + 15 from Provider 2)
- Pro Tier Models: 1 model (Gemini 2.5 Pro)
- Ultra Tier Exclusive: 1 model (Gemini 3 Pro Preview)
Important Notes
- Automatic Access: All free tier promotions are effective immediately. No action required from users to access newly available models.
- Provider 2 Transition: Legacy Provider 2 models discontinued due to provider bankruptcy. New Provider 2 infrastructure offers superior stability and expanded model catalog.
- Tier Benefits: While 24 models are now free, Pro and Ultra tiers continue to provide access to cutting-edge flagships like Gemini 2.5 Pro and Gemini 3 Pro Preview, along with enhanced rate limits and priority support.