Mistral AI closed a €3 billion Series D funding round on September 8, 2026. Samsung Electronics led this round, with Scaleup Europe Fund and PSG Equity acting as co-leads. This capital injection pushes the company’s valuation beyond €21 billion, which is a significant increase from the €11.7 billion valuation from its 2025 Series C. The funding follows an 830 million dollar loan the company took in March to finance its own data centers. New investors in this round include Advent, funds managed by BlackRock, and the Grand Duchy of Luxembourg. Previous backers like NVIDIA, ASML, and a16z also participated. Mistral AI currently operates in 20 countries and serves more than 125 companies, including Airbus, ASML, and HSBC.
Mistral AI pursues a strategy that focuses on the enterprise sector through open-weight models and proprietary infrastructure. The company aims to provide an alternative to the closed platforms controlled by a few US providers. While Mistral’s models, such as Medium 3.5, do not currently rank at the top in terms of performance, trailing behind Chinese competitors like Qwen and Kimi, the company focuses on profitable niches and sovereign AI. This strategy involves building local know-how in the European ecosystem and providing models that customers can control. The company expects to reach approximately $1 billion in annual recurring revenue by the end of 2026.
The three distinct billing streams of Mistral
Mistral creates a management challenge for finance teams because it separates billing into three completely different systems. A single team can generate costs across these three streams simultaneously. The first stream is the Mistral API, which uses a pay-per-token model through la Plateforme. The second stream is Le Chat, which is a subscription-based consumer and business chat interface. The third stream involves self-hosted models where costs consist of GPU compute on platforms like AWS or GCP.
| Le Chat Plan | Monthly Price | Key Capabilities |
|---|---|---|
| Free | $0 | 25 messages/day, 40+ connectors, 500 memories |
| Pro | $14.99 | 150 Flash Answers, 15GB storage, Mistral Vibe |
| Team | $19.99 (Annual) | 30GB/user, admin controls, shared RAG library |
| Enterprise | Custom | On-prem, custom models, SAML SSO, audit logs |
The Mistral Pro subscription costs $14.99 per month and provides users with access to SOTA models, 150 Flash Answers per day, 15GB of storage, and the ability to use Mistral Vibe for in-chat coding tasks. You should examine the specific API rate limits if you intend to move a production workload to Mistral’s la Plateforme. A Pro subscription does not cover API calls, and API credits do not apply to Le Chat. This separation means that the API and the subscription function as two different products with different invoices. For teams, the Team plan costs $24.99 per user per month with monthly billing or $19.99 per user per month with annual billing.
Cohere’s strategy for enterprise search and RAG
Cohere focuses its enterprise bet on retrieval-augmented generation (RAG), embeddings, and enterprise search. Its Command R+ model is designed specifically for retrieval at enterprise scale. This approach targets organizations with large internal document corpora that require specialized knowledge search. Cohere’s strategy differs from Mistral’s broader focus on open-weight availability and infrastructure.
| Cohere Model | Input $/1M tokens | Output $/1M tokens | Context Window |
|---|---|---|---|
| Command R+ (08-2024) | $2.50 | $10.00 | 128K |
| Command A+ | $0.30 | $1.50 | 192K |
| Command A | $2.50 | $10.00 | 256K |
| Command R (08-2024) | $0.15 | $0.60 | 128K |
| Command R7B (12-2024) | $0.04 | $0.15 | 128K |
Cohere uses Alice to improve model safety and reliability. Alice provides red teaming insights and targeted data to the AI safety team. This process allows the company to develop more sophisticated safety mechanisms and accelerate model release timelines. The AI safety team uses findings from Alice to identify necessary focus areas for mitigations. This methodology helps prevent the generation of harmful content in high-risk areas such as misinformation and hate speech.
Comparing intelligence and reasoning benchmarks
Performance metrics show different strengths for Mistral and Cohere models across various tasks. Mistral provides high performance in coding and math, while Cohere shows high intelligence in certain reasoning tasks.
| Benchmark | Cohere (Command R+) | Mistral AI (Nemo/Medium) |
|---|---|---|
| Intelligence | 33.6 | 25.0 |
| Coding | 27.8 | 46.9 |
| Math | 13.0 | 40.3 |
| MMLU Pro | 71.2 | 76.2 |
| GPQA | 76.1 | 74.8 |
| LiveCodeBench | 28.7 | 52.7 |
| Aider | 38.3 | 65.4 |
| AIME | 9.7 | 70.0 |
| BBH | 42.8 | 36.1 |
Mistral Large 2411 achieves an Aider score of 65.4, which is significantly higher than the 38.3 score for Cohere’s Command R. Magistral Medium shows high performance in math with a score of 40.3 and a LiveCodeBench score of 52.7. In contrast, Cohere’s Command A+ shows an intelligence score of 33.6. Mistral Nemo has an intelligence score of 25.0. These numbers show that Mistral holds an advantage in coding and math, while Cohere performs well in intelligence and GPQA tasks. Will the massive capital injection from Samsung successfully bridge the performance gap between Mistral and US-based frontier models?
The legal importance of European data residency
Data residency is a requirement for many European enterprises due to the General Data Protection Regulation (GDPR). Compliance with GDPR is a legal obligation that involves strict requirements for how the company collects, processes, and stores personal data of EU residents. While GDPR does not mandate data residency within the EU, it places significant restrictions on transferring data outside the European Economic Area. Non-compliance can result in fines of up to 4% of a company’s global annual revenue or €20 million.
Mistral provides a solution for these requirements through its focus on data residency and sovereignty. The company allows for data residency, on-premise deployment, and deployment in a private cloud. This ensures that data stays within the organization’s own environment. Mistral’s French jurisdiction provides an additional layer of legal certainty for European firms.
Cohere also provides private deployment options for US defense and government clients. Both companies offer the ability to run models on-premises to satisfy air-gapped requirements. However, Mistral’s focus on European sovereignty and its identity as a European firm provide a direct alignment with the needs of EU-based enterprises. For international companies, setting up local data centers is a way to reduce latency and ensure compliance with local laws.
Cost analysis for high-volume production models
Production AI features require cost-effective inference to remain profitable. Mistral undercuts major US providers on input and output costs for its flagship models. Mistral Large 3 at $0.50 per million input tokens and $1.50 per million output tokens is 80% cheaper than GPT-5.4 on input and 90% cheaper on output. Against Claude Sonnet 4.6, it is 83% cheaper on input and 90% cheaper on output.
| Model | Input $/1M tokens | Output $/1M tokens |
|---|---|---|
| Mistral Large 3 | $0.50 | $1.50 |
| Mistral Small 4 | $0.15 | $0.60 |
| Mistral Nemo | $0.02 | $0.03 |
| Cohere Command R+ | $2.50 | $10.00 |
| Cohere Command R | $0.15 | $0.60 |
Mistral Small 4 competes with Gemini 2.5 Flash, providing half the price on input and one-quarter on output. For high-throughput classification and extraction, Small 4 is a cost-effective option. Mistral Nemo is the cheapest model in the lineup at $0.02 for input and $0.03 for output per million tokens. Cohere’s Command R provides a price of $0.15 for input and $0.60 for output per million tokens.
Organizations running high-volume workloads can also choose to self-host Mistral’s open-weight models. This approach results in zero per-token fees, though the company must pay for GPU compute. For workloads exceeding tens of millions of tokens per day, self-hosting becomes more economical than the API.
Mistral Le Chat service tiers
Mistral provides different tiers for Le Chat to accommodate various user needs. The Free tier allows users to access SOTA models, 40+ connectors, and image generation with a soft cap of roughly 25 messages per day. The Pro tier costs $14.99 per month and includes the Mistral Vibe coding workspace. The Team tier costs $24.99 per user per month with monthly billing or $19.99 per user per month with annual billing.
| Feature | Free | Pro | Team |
|---|---|---|---|
| Messages | ~25/day soft cap | 6x multiplier | 6x multiplier |
| Storage | 500 memories | 15GB | 30GB/user |
| Coding | Basic | Mistral Vibe | Mistral Vibe |
| Admin Controls | None | None | Admin API, SSO |
The Team plan includes domain name verification and centralized billing. It also provides a shared knowledge and RAG library for team-wide document access. The Enterprise tier is custom-priced and starts at approximately $20,000 per month. This tier provides private or on-premise deployment, custom models fine-tuned on organizational data, and full audit logs for compliance reporting. Large organizations in regulated industries like finance and healthcare use the Enterprise tier for privacy-sensitive workflows.
The verdict on the EU enterprise LLM race
Mistral wins the EU enterprise race for companies that prioritize strict data residency, low-cost production inference, and the ability to host models on-premises. Its massive funding from Samsung and its focus on a full-stack AI platform give it the resources to compete on infrastructure. Mistral provides the most flexible options for developers who need to manage both API calls and subscription-based chat tools.
Cohere wins for enterprises that require specialized RAG capabilities and a proven track record in the US government sector. Its models are built for high-scale retrieval and search tasks within large internal document sets. The choice between the two depends on whether a firm needs a specialized retrieval tool or a sovereign, customizable infrastructure. Mistral is the stronger choice for companies that want to avoid vendor lock-in through open-weight models.
