More categories More architectures Solid LLaMA 3 update, reliable for coding, chat, and Q&A tasks A plain shell sandbox on DHI's shell-docker template: bash, git, apt, a Docker engine with compos... llmman with llama.cpp: server, server-cuda, server-cuda13, server-rocm, server-vulkan Base image for the Shell sandbox kit for Docker Sandboxes Distilled LLaMA by DeepSeek, fast and optimized for real-world tasks Anthropic's Claude Code, with API-key and subscription sign-in both resolved by the sandbox proxy. Versatile Qwen update with better language skills and wider support SmolLM3 is a 3.1B model for efficient on-device use, with strong performance in chat

Google’s latest Gemma, in its QAT (quantization aware trained) variant Newest LLama 3 release with improved reasoning and generation quality Microsoft’s compact model, surprisingly capable at reasoning and code Ministral 3: compact vision-enabled model with near-24B performance, optimized for local edge use Efficient 80B MoE coding model with 3B activated params, 256K context, and agentic capabilities Efficient open model with top-tier performance and fast inference Gemma 4: multimodal open AI models by Google, optimized for reasoning, coding, and long context. DeepCoder-14B-Preview is a code reasoning LLM fine-tuned to scale up to long context lengths Kimi K2 Thinking: open-source agent with deep reasoning, stable tool use, fast INT4, 256k context.