Timestral vs. Mistral Large 3

Timestral is a comprehensive suite of time-series foundation models accompanied by a Python inference toolkit, specifically designed to enable the creation of multivariate probabilistic forecasts derived from historical data sequences. It offers four distinct model sizes, including 5M, 28M, 123M, and 209M parameters, each capable of managing past-only, known-future, and categorical covariates, while providing quantile forecasts that can accommodate context windows of up to 32,768 time steps. Users can conveniently load pretrained weights from the Hugging Face Hub or from their local files, allowing for inference on both CPU and GPU setups. These forecasts can be integrated effortlessly into a wide array of applications, such as monitoring systems, capacity planning, energy management, supply chain logistics, financial risk analysis, and maintenance operations. Additionally, Timestral boasts zero-shot forecasting capabilities and can be customized to specific domain data using the Mistral Forge, further enhancing its versatility. This toolkit primarily targets data scientists, machine learning engineers, and collaborative teams eager to devise effective forecasting solutions; its repository code is available under the Apache 2.0 license, while the model weights are managed by a distinct Modified MIT license. With its robust and comprehensive features, Timestral is set to considerably improve the precision and effectiveness of predictive analytics across various sectors, ensuring that users can make informed decisions based on reliable forecasts. As industries increasingly rely on data-driven insights, Timestral stands out as an essential resource for achieving superior forecasting outcomes.

Mistral Large 3 is a frontier-scale open AI model built on a sophisticated Mixture-of-Experts framework that unlocks 41B active parameters per step while maintaining a massive 675B total parameter capacity. This architecture lets the model deliver exceptional reasoning, multilingual mastery, and multimodal understanding at a fraction of the compute cost typically associated with models of this scale. Trained entirely from scratch on 3,000 NVIDIA H200 GPUs, it reaches competitive alignment performance with leading closed models, while achieving best-in-class results among permissively licensed alternatives. Mistral Large 3 includes base and instruction editions, supports images natively, and will soon introduce a reasoning-optimized version capable of even deeper thought chains. Its inference stack has been carefully co-designed with NVIDIA, enabling efficient low-precision execution, optimized MoE kernels, speculative decoding, and smooth long-context handling on Blackwell NVL72 systems and enterprise-grade clusters. Through collaborations with vLLM and Red Hat, developers gain an easy path to run Large 3 on single-node 8×A100 or 8×H100 environments with strong throughput and stability. The model is available across Mistral AI Studio, Amazon Bedrock, Azure Foundry, Hugging Face, Fireworks, OpenRouter, Modal, and more, ensuring turnkey access for development teams. Enterprises can go further with Mistral’s custom-training program, tailoring the model to proprietary data, regulatory workflows, or industry-specific tasks. From agentic applications to multilingual customer automation, creative workflows, edge deployment, and advanced tool-use systems, Mistral Large 3 adapts to a wide range of production scenarios. With this release, Mistral positions the 3-series as a complete family—spanning lightweight edge models to frontier-scale MoE intelligence—while remaining fully open, customizable, and performance-optimized across the stack.