Mistral Forge

What it does

Forge builds customized large language models from an organization's own material, including unstructured documents, domain terminology, ontologies, decision frameworks, codebases, telemetry and incident timelines. It trains, aligns and evaluates models on that data. The output is a model version, with the datasets, training runs and configurations tracked as versioned assets and served through an inference runtime that can enforce internal governance and safety constraints.

Teams use Forge across the model lifecycle. It can generate synthetic domain and edge-case training samples, run pre-training or post-training with methods such as SFT, DPO, RLHF and LoRA, and evaluate models against enterprise KPIs with regression and drift checks. If a new version underperforms, teams can roll back to a known-good one. Named applications include code modernization, industrial domain adaptation, cybersecurity investigation and quant research. Its emphasis is auditability and deployment control rather than a general chat interface.

Forge suits organizations with large proprietary datasets and strict data isolation or compliance needs. Models can be deployed in private cloud, on-premises or on Mistral compute, with control over data residency. Access starts with a 'Talk to an expert' conversation rather than self-serve sign-up. Request a quote from Mistral for pricing, trial options and supported data connectors.

What it can do

• Generate synthetic domain examples and edge-case training samples from enterprise workflows.
• Train models on large volumes of unstructured enterprise data with Dense or Mixture-of-Experts architectures.
• Align models with SFT, DPO, RLHF and LoRA-based adaptation.
• Evaluate models against enterprise KPIs with regression suites and drift detection.
• Version models, datasets, training runs and configs with rollback and audit lineage.
• Serve customized models with policy-aware inference across private cloud, on-prem or Mistral compute.

Example workflows

Domain adaptation of an engineering assistant

Illustrative example. Input: engineering documentation, standards, vocabularies and decision frameworks. Steps: prepare domain examples, train on the unstructured corpus, align with SFT or DPO, and evaluate against internal KPIs. Output: a versioned customized model with evaluation results and lineage for each training run.

Legacy code modernization review

Illustrative example. Input: proprietary codebases and engineering standards. Steps: train on the codebase, align to architecture practices, and run regression suites after each version change. Output: a model that proposes reviewable refactors and migrations, with rollback to a prior version if regressions appear.

Before you choose

• Forge is an enterprise custom-training product with a sales-led path; request a quote for pricing and commercial terms.
• Customized models need evaluation against your own KPIs and drift monitoring. Results depend on your data quality and governance.
• Forge is separate from Mistral's chat, Vibe and API products, so their price pages do not apply to it.

Listed under

• Coding
• Sandboxes and infrastructure
• Compliance
• Custom LLM training
• Custom models
• Data governance
• Enterprise
• Enterprise AI

Get started

The page's primary call to action is 'Talk to an expert'; no self-serve sign-up for Forge is shown.

1. Contact Mistral through the 'Talk to an expert' route on the Forge page to scope the project.
2. Identify the proprietary datasets, ontologies and policies to be used for training, and confirm data isolation requirements.
3. Choose the deployment environment (private cloud, on-prem or Mistral compute) that matches your risk and data residency needs.

Contact Mistral about the model-building engagement and deployment terms. Vibe subscriptions and Studio API prices are separate offerings.

We couldn’t find published prices for Mistral Forge. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Mistral Forge - Build AI models that know your enterprise ↗mistral.ai
• Pricing | Mistral ↗mistral.ai

Company website

mistral.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.