Models

⚡ The Brief Llama 4 ships in two MoE models only: Scout with sixteen experts and Maverick with one hundred twenty-eight experts, both activating seventeen billion parameters. Scout delivers a ten million token context window, the longest in any production open-weight model as of April 2026. Maverick uses four hundred…

⚡ The Brief DeepSeek V4-Pro and V4-Flash are open-weight Mixture-of-Experts (MoE) models released on April 27, 2026, featuring a commercially permissive MIT-derived license enabling production deployment. V4-Pro is designed for frontier-class benchmarks emphasizing maximum quality, while V4-Flash prioritizes lower latency and cost-efficiency with smaller active parameters during inference. Multiple self-hosting…

⚡ The Brief Qwen 3.5 spans 0.8B to 397B parameters with eight models including three Mixture-of-Experts variants under Apache 2.0 license. Qwen3.5-9B leads downloads at 7.2M, while flagship Qwen3.5-397B-A17B activates only 17B per token despite 397B total capacity. Qwen 3.6 introduces refreshed 27B dense and 35B-A3B MoE with improved multilingual…

⚡ The Brief Mistral Large 3 (675B dense flagship) and Small 4 (119B, released April 2026) anchor the 2026 production lineup for diverse AI workloads. Devstral Small 2 (24B) leads in downloads with over 308,000, highlighting the community’s focus on specialized coding models for self-hosting. Ministral 3 provides on-device inference…

⚡ The Brief Gemma 4 ships in four configurations from 2B edge models to a 31B dense flagship, all open-weight under custom commercial license. Phi-4 family spans mini variants, reasoning specialists, and a 15B vision model, all MIT-licensed for unrestricted commercial deployment. Both families target on-device inference and cost-efficient cloud…