French AI lab Mistral made a bold play for enterprise adoption this week, unveiling two major products at NVIDIA GTC 2026: the Forge platform for custom model training and Mistral Small 4, a powerful open-source model released under the Apache 2.0 license.
Forge: Build-Your-Own AI for the Enterprise
Mistral Forge is a platform designed to let enterprises and governments customize AI models using their own proprietary data. Rather than relying on general-purpose models, organizations can fine-tune Mistral's foundation models for their specific domains and workflows.
The platform is already in use by a notable roster of early adopters. Ericsson, the European Space Agency, Italian consulting firm Reply, and Singapore's DSO and HTX are among the first partners. ASML, the Dutch chipmaker that led Mistral's Series C round last September at an €11.7 billion valuation, is also an early adopter.
The move positions Mistral directly against OpenAI and Anthropic in the enterprise AI market, with a differentiated pitch: rather than offering API access to a shared model, Forge gives organizations the tools to build something that is genuinely their own.
Small 4: Open Source With Serious Specs
Alongside Forge, Mistral released Small 4 — a mixture-of-experts model with 119 billion total parameters but only 6 billion active per query. The architecture delivers meaningful efficiency gains: 40 percent faster inference than its predecessor with three times higher throughput.
Small 4 supports a 256k context window for handling long-form documents and conversations. It also introduces configurable reasoning, letting developers control how much compute the model dedicates to chain-of-thought processing depending on the task.
The model unifies capabilities that were previously spread across separate Mistral releases — the instruction-following of Mistral Small, reasoning from Magistral, multimodal features from Pixtral, and agentic coding from Devstral — into a single model.
Why It Matters
Mistral's dual announcement reflects a broader strategic shift in the AI industry. On one side, enterprises want models trained on their own data for compliance, accuracy, and competitive advantage. On the other, the open-source community continues to demand capable models without restrictive licenses.
By releasing Small 4 under Apache 2.0 while simultaneously launching a premium enterprise platform, Mistral is betting it can serve both audiences. The timing — during NVIDIA's marquee annual conference — ensures maximum visibility among the infrastructure buyers and enterprise decision-makers who fill GTC's halls.
With major cloud providers and enterprise software companies increasingly bundling AI into their platforms, Mistral's independent, Europe-based approach offers an alternative for organizations wary of vendor lock-in with American hyperscalers.
