Mistral releases Large 4 with a score of 38 on the Intelligence Index

Mistral releases Large 4 with a score of 38 on the Intelligence Index

Mistral has released Mistral Large 4 in Research Public Preview. The model scores 38 on the Artificial Analysis Intelligence Index, comparable to GPT-6 Luna (max, 38) and DeepSeek V4.1 Flash (max, 39). The result places France among the countries with the highest-scoring AI models outside the United States and China.

Mistral plans to release the weights for the model at the end of October. Large 4 has 1 trillion parameters, with 49 billion active parameters.

Benchmark results

Intelligence Index

Mistral Large 4 Preview scores 38 on the Artificial Analysis Intelligence Index. This is comparable to GPT-6 Luna (max, 38) and close to DeepSeek V4.1 Flash (max, 39). According to the reported rankings, it is the highest-scoring model from outside the United States and China, ahead of models from countries including South Korea and the United Arab Emirates.

Cyber Index

The model scores 50 on the Artificial Analysis Cyber Index, matching GLM-5.3-Flash and ranking ahead of models such as Kimi K3 and DeepSeek V4.1 Flash (max). Its strongest result is on CyberGym-E2E-AA, where it scores 82%. That result is higher than MiMo-V2.6-Pro (79%) and GPT-6 Luna (max, 78%).

Once the weights are released, Mistral Large 4 is expected to rank among the top three open-weights models on the Cyber Index. It remains behind MiMo-V2.6-Pro, which scores 56 on the index.

Cost per task

Mistral Large 4 Preview costs $1.13 per Intelligence Index task at standard pricing. Its token prices are:

• Input: $1.36 per 1 million tokens
• Output: $4.18 per 1 million tokens
• Cached input: $0.14 per 1 million tokens

For the first two weeks, the model is available with a 50% launch discount. This reduces the estimated cost per task to $0.57, with discounted token prices of $0.68 per 1 million input tokens and $2.09 per 1 million output tokens.

The discounted cost remains higher than the reported costs for GLM-5.3-Flash ($0.25) and DeepSeek V4.1 Flash (max, $0.27).

Document and image reasoning

On GDP.pdf, Mistral Large 4 Preview scores 19%, matching MiMo-V2.6-Pro (19%) and trailing Kimi K3 (22%). This represents an 18-point improvement over Mistral Large 3.

The API now accepts up to 100 images per request, compared with eight images for previous Mistral models. The expanded image limit contributes to the model's document and image-processing capabilities.

Model specifications

• Context window: 512,000 tokens
• Input modalities: Text and images
• Output modality: Text
• Standard pricing: $1.36 per 1 million input tokens and $4.18 per 1 million output tokens
• Cached input pricing: $0.14 per 1 million tokens
• Launch pricing: 50% off for the first two weeks, or $0.68 per 1 million input tokens and $2.09 per 1 million output tokens
• Availability: Research Public Preview through Mistral's API
• Weights: Planned for release at the end of October