Hermes 3 is a generalist language model with many improvements over Hermes 2, including advanced agentic capabilities, better roleplaying, reasoning, multi-turn conversation, and long context coherence. This 70B model is a competitive finetune of Llama-3.1-70B focused on aligning LLMs to the user with powerful steering capabilities.

MiroThinker is the official implementation of the MiroMind Research Agent Project. It is an open-source search agent designed to advance tool-augmented reasoning and information-seeking capabilities, enabling complex real-world research workflows across diverse challenges.

Gemini 3 Pro with native image generation support. Can generate images inline via text prompts.

Cydonia 24B v4.3 continues TheDrummer's Cydonia series with updated tuning on Mistral Small.

Magidonia 24B v4.3 is a new 24B Drummer finetune built for rich, creative roleplay.

Derestricted GLM 4.6 tuned for open-ended creative writing and roleplay with relaxed filters.

GLM-4.7 is a next-gen GLM series text model with stronger reasoning, long-context chat, and reliable tool use. Routed directly via Z-AI (Zhipu).

GLM-4.7 original with extended thinking capabilities for complex reasoning.

GLM-4.7 is a next-gen GLM series text model with stronger reasoning, long-context chat, and reliable tool use.

GLM-4.7 with extended thinking capabilities for enhanced reasoning on complex tasks.

Lightweight tier optimized for speed and cost.

The flagship model for balanced reasoning and context-rich dialogue. Perfect for AI roleplay, storytelling, and assistant tasks with a 16K window.

Tailored for deep reasoning, long-form generation, and RAG workloads. 32K token context window.

MiniMax M2.1 builds on M2 with enhanced context understanding and improved complex tool use. 230B parameter MoE model (10B active) optimized for agentic workflows and long-horizon tasks.

Google's Gemini 3 Flash preview model optimized for speed while maintaining high capability. Features sub-second response times with strong multimodal understanding and reasoning.

Google's Gemini 3 Flash preview model with thinking mode enabled for enhanced reasoning and chain-of-thought capabilities.

Mistral Small Creative is an experimental model designed for creative writing, narrative generation, roleplay and character-driven dialogue, general-purpose instruction following, and conversational agents.

Nvidia's latest Nemotron 3 Nano model with 30B total parameters (3B active) using hybrid Mamba-Transformer MoE architecture. Features excellent throughput and strong reasoning capabilities.

Essential AI's RNJ-1 Instruct 8B model. A capable instruction-following model optimized for general chat and task completion.

ReExtract variant of the Iceblink LoRA for GLM 4.5 Air with refined extractions for creative writing.

ReExtract variant of the v2 Iceblink LoRA for GLM 4.5 Air with enhanced creative extractions.

ReExtract variant of the Steam LoRA for GLM 4.5 Air with refined uncensored extractions for creative RP.

Mixtral 8x7B is a high-quality sparse Mixture of Experts (MoE) model with 45B total parameters but only 13B active per token. Excels at text generation, summarization, question answering, and code generation. Supports English, French, German, Spanish, and Italian. Apache 2.0 licensed.

Mixtral 8x22B is a powerful sparse Mixture of Experts (MoE) model with 141B total parameters and 39B active per token. Features a 64K context window, exceptional math performance, and cost-efficient inference. Supports English, French, German, Spanish, and Italian. Apache 2.0 licensed.

GLM-4.6, Zhipu's flagship text model with 256K context window and advanced reasoning capabilities. Direct via Z-AI (Zhipu).

GLM-4.6V scales its context window to 128k tokens in training, and achieves SoTA performance in visual understanding among models of similar parameter scales. Integrates native Function Calling capabilities, bridging 'visual perception' and 'executable action' for multimodal agents. Quantized at FP8.

Nex-AGI's flagship 671B agentic model built on DeepSeek V3.1. Optimized for programming, tool use, web search, multi-hop reasoning, and mini-app development. Industry-leading results on SWE-bench Verified (70.6%), τ2-Bench (80.2%), and BFCL v4 (65.3%). Full production-ready agent capabilities with 128K context.

Devstral 2 123B is a 123 billion parameter model from Mistral AI optimized for coding and development tasks. Features advanced reasoning capabilities for software engineering workflows.

GLM-4.6V-Flash (9B), a lightweight model optimized for local deployment and low-latency applications. Scales context window to 128k tokens and achieves SoTA performance in visual understanding among similar-scale models.

GLM-4.6V scales its context window to 128k tokens in training, and achieves SoTA performance in visual understanding among models of similar parameter scales. Integrates native Function Calling capabilities, bridging 'visual perception' and 'executable action' for multimodal agents. Direct via Z-AI (Zhipu).