Mistral Small 4

条件：Set on 2026-10-20 from the model card: agentic capabilities with native function calling and JSON output, image input support, and a togglable reasoning mode.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Re-verified on 2026-10-20 from the model card: 256k context length.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Re-verified on 2026-10-20; the card is tagged vLLM and provides NVFP4 and eagle-head checkpoints. The minimum 4x H100 footprint was recorded earlier from the release announcement and is not restated on the model card.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Set on 2026-10-20 because the mistralai Hugging Face repository for Small 4 remains public, not gated and not disabled; Mistral does not publish a formal lifecycle table for these weights, so this reflects current availability rather than a vendor-stated status.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Re-verified on 2026-10-20; the card states Apache 2.0 for commercial and non-commercial use and front matter declares apache-2.0.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Set on 2026-10-20 from the official repository name suffix 2603; the model card does not state a day, so only month precision is asserted here.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Re-verified on 2026-10-20 from the model card: 119B parameters with 6.5B activated per token.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.

条件：Re-verified on 2026-10-20 against the official Mistral AI Hugging Face repository, which publishes the Small 4 weights, NVFP4 quantisation and an eagle head and is not gated.

证据：Mistral Small 4 is a hybrid model that unifies Instruct, Reasoning and Devstral capabilities into one multimodal model. Key features: MoE with 128 experts and 4 active; 119B parameters with 6.5B activated per token; 256k context length; multimodal input accepting text and images with text output; instruct and reasoning functionality with function calls and configurable reasoning effort; vision, multilingual and agentic capabilities with native function calling and JSON output; Apache 2.0 license for commercial and non-commercial use. The card is tagged vLLM and links NVFP4 and eagle speculative-decoding checkpoints. Front matter: license: apache-2.0. The Hugging Face model API reports the repository as public and not gated.