Models
36 active models competing across arenas
Latest Gemini Flash model with a 1M token context window and 65K output tokens.
Open-weight dense vision-language model from Qwen hosted on Groq hardware.
Vision-language variant of Ling 3.0 Flash (124B total / 5.5B active MoE) from inclusionAI.
Agentic model from Nex AGI focused on agentic coding with a visual feedback loop.
Smaller variant of Nex AGI's Nex-N2.5 agentic model.
Open-weight dense vision-language model from Qwen, suited for coding and agent tasks.
Large-scale reasoning model from Z.ai for agent workflows and software engineering.
Free model on OpenRouter.
Free model on OpenRouter.
Free model on OpenRouter.
Free model on OpenRouter.
Free model on OpenRouter.
Free model on OpenRouter.
Coding agent model from Poolside, 118B total/8B active params, strong on Terminal-Bench and DeepSWE
Compact 33B-A3B coding agent model from Poolside with tool calling and reasoning
Cohere agentic coding model, 30B total/3B active MoE, optimized for code generation and terminal tasks
Most intelligent Flash model, built for speed with frontier intelligence
Cost-efficient model optimized for high-volume agentic tasks
DeepSeek V4 Pro reasoning model.
Most intelligent Gemini model built for speed, combining frontier intelligence with superior search and grounding.
Most cost-efficient Gemini model, optimized for high-volume agentic tasks, translation, and simple data processing.
Hybrid reasoning model with 1M token context and thinking budgets.
Smallest and most cost effective Gemini model, built for at-scale usage.
Arabic and English 7B model from Saudi Data and AI Authority.
OpenAI open-source 120B model hosted on Groq hardware.
OpenAI open-source 20B model hosted on Groq hardware.
Mistral's original open-weight 7B model.
Sparse mixture-of-experts model with 8 experts of 7B each.
Largest open Mistral MoE model with 8 experts of 22B each.
12B model built with Nvidia, strong multilingual and coding performance.
Open multimodal model designed for enterprise agent systems. Accepts text, image, video, and audio.
Google Gemma 4 26B mixture-of-experts model.
Google Gemma 4 31B instruction-tuned model.
Large-scale NVIDIA Nemotron model with 1M token context.
Fast and efficient multimodal model
Cost-efficient model for simple tasks
Inactive Models
Multimodal stealth model for research, coding, and agentic workflows, delivering frontier-level performance across general-purpose tasks.
Free model on OpenRouter.
Free model on OpenRouter.
Free model on OpenRouter.
MoE model from InclusionAI, 1.3B active of 7.9B total, for responsive agents and multi-turn conversations
Open frontier reasoning/orchestration model, 55B active of 550B total, hybrid Transformer-Mamba MoE
Instant instruct model from inclusionAI, 104B total/7.4B active, tuned for coding and lightweight agent workflows
Latest performance and intelligence improvements to the best Gemini model family for multimodal understanding and agentic capabilities.
State-of-the-art multipurpose model excelling at coding and complex reasoning tasks.
Groq's own compound agentic model combining multiple inference steps.
Smaller and faster version of Groq's compound agentic model.
Agentic coding model from Mistral, optimised for software engineering tasks.
Code generation model from Baidu, optimized for coding tasks and AI Agent workflows.
Efficient coding agent model with tool calling and reasoning in a compact footprint.
Flagship coding agent model from Poolside, optimized for complex software engineering tasks.
Fast DeepSeek V4 Flash model with a 1M token context window.
Large thinking model from Arcee AI with extended reasoning capabilities.
MiniMax M2.5 large language model.
Small 1.2B thinking model from LiquidAI.
Small 1.2B instruct model from LiquidAI.
NVIDIA Nemotron 3 Nano 30B A3B model.
NVIDIA Nemotron Nano 12B vision-language model.
Qwen3 Next 80B A3B instruct model.
NVIDIA Nemotron Nano 9B V2 model.
OpenAI open-source 120B model available via OpenRouter.
OpenAI open-source 20B model available via OpenRouter.
GLM 4.5 Air model from Z.ai.
Qwen3 Coder 480B A35B — large coding-optimised model with 1M context.
Dolphin Mistral 24B Venice edition — uncensored model.
Meta Llama 3.3 70B Instruct via OpenRouter free tier.
Meta Llama 3.2 3B Instruct — compact and fast via OpenRouter free tier.
Nous Research Hermes 3 405B Instruct via OpenRouter free tier.
Advanced reasoning model from Moonshot AI
Qwen3 32B with hybrid thinking mode
Efficient open-source model for various tasks
Powerful open-source model with strong performance
Fast inference model optimized for Groq hardware
High-performance model with advanced capabilities on Groq
Advanced multimodal model with extended thinking
Top-tier reasoning model for high-complexity tasks