Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Mercor raises $750 million at $20 billion valuation to expand AI train - OpenSmartRoute
Mercor raises $750 million at $20 billion valuation to expand AI training
Mercor announced a $750 million funding round, valuing the company at $20 billion. The money will boost infrastructure and research for AI model evaluation and training.
Key points
Mercor's Series D funding raised $750 million on November 10, 2026
The valuation of Mercor reached $20 billion after the funding
The funding included participation from NVIDIA, Salesforce, Zoom, and others
Mercor opened offices in New York and London in 2026
Why it matters: Cost reduction and improved performance in AI training and evaluation can save time and money for model developers.
By OpenSmartRoute editorial · written through the router by llm-small
From Unite.AI - “Mercor Raises $750M Series D at $20B Valuation to Expand AI Training”
Introduction - what Mercor announced and why it matters
Mercor announced a new funding round on November 10, 2026. The company raised 750 million dollars in a Series D investment. This funding values Mercor at 20 billion dollars. The company focuses on training and evaluating artificial intelligence (AI) models. This news is important because it shows how much investment is going into AI infrastructure and research. It also highlights Mercor’s role in improving AI performance in real-world settings.
This funding will help Mercor expand its work with enterprise customers. It aims to develop better tools for testing AI models on actual workflows. The company also plans to improve the environments where agents learn complex tasks. These steps could speed up AI development and make models more useful for businesses. The investment signals confidence in Mercor’s approach and its potential to shape AI training methods.
Details of the funding round - amount, participants, and valuation
Mercor’s latest funding round raised 750 million dollars. Several well-known companies and investment firms participated. These include NVIDIA, Salesforce, and Zoom. Other investors include Benchmark, Felicis, Fundomo, and funds managed by affiliates of Apollo. Additional participants are G Squared, General Catalyst, Glade Brook Capital, Sapphire Ventures, StepStone Group, and XTX Ventures.
The total valuation of Mercor after this round is 20 billion dollars. This makes it one of the most highly valued companies in AI training. The funding round was a Series D, which is usually for companies that have already shown growth and are expanding further. The large amount of money reflects strong investor confidence. It also indicates the importance of AI training infrastructure in the current market.
Purpose of the funding - infrastructure, research, and environment expansion
Mercor plans to use the new funds to build more infrastructure. This includes tools for evaluating AI models and improving them after initial training. The company wants to create more environments where agents can learn complex workflows. These environments simulate real-world tasks, helping models perform better in actual use cases.
Pierre Fabre and XtalPi announced a partnership to discover new small-molecule drugs. XtalPi provides AI and robotics, while Pierre Fabre handles selection and development.
The book explores why employees resist adopting AI fully and how leaders can encourage organization-wide AI use.
The funding will also support research efforts. Mercor aims to develop new methods for measuring how well models work. It will focus on reducing the resources needed for training and post-training. This means making AI models more efficient and less costly to improve. The company also wants to expand its physical presence by opening new offices. In 2026, Mercor opened offices in New York and London. These locations are closer to enterprise customers and research labs. The goal is to better serve clients and attract talent.
Mercor's enterprise evaluation approach - using real workflows and experts
Mercor has spent the past two years creating infrastructure to test AI models on real-world work. It uses human experts to evaluate how models perform in practical tasks. These evaluations are designed to see what AI agents can do, how much they cost, and how their performance can improve over time.
The company’s approach is to use actual workflows from businesses. It measures models against specific company requirements. This helps identify where models need more training or practice. Mercor’s evaluations are tailored to each enterprise, making the results more relevant. This method contrasts with testing models on generic tasks, providing a clearer picture of real-world usefulness.
Mercor reports that it has a network of over five million experts. These experts are paid to help train AI models. The company notes that in the United States alone, there are more than 800 different occupations with workflows that models have yet to be evaluated against. Its enterprise playbook starts with an open model, then builds public evaluations. It uses post-training to cut costs before training models specifically for customers.
Expert network and evaluation methods - measuring model capabilities and costs
Mercor’s evaluation system relies on a large network of human experts. These experts create tasks and verifiers that translate human judgment into signals for training models. The tasks are based on real workflows, and verifiers determine whether the model succeeded. This process helps measure how well models perform on economically valuable tasks.
The company uses a system called APEX to assess model performance. APEX measures how models do on real-world tasks across thousands of workflows. It provides a way to see how models handle different types of work and how much they cost to run. This helps companies understand the value of their AI models and where to improve them.
Mercor’s approach focuses on both capabilities and costs. By measuring these factors, companies can decide whether a model is ready for deployment. They can also identify areas where additional training or resources are needed. This comprehensive evaluation helps optimize AI performance in practical settings.
Post-training and model improvement - reducing training resources
After initial training, Mercor uses post-training methods to improve models further. Post-training involves additional learning on specific tasks or workflows. This process requires fewer resources than training a model from scratch.
Mercor reports that post-training can use only a fraction of the resources needed for frontier training. Frontier training refers to the initial development of a model. Post-training makes models more accurate and efficient without the high costs of full retraining. This approach allows companies to update their models more quickly and at lower expense.
The company’s infrastructure supports this process by providing environments where models can practice and learn from real workflows. These environments are recreated using software from the acquisition of Deeptune. They include tasks, verifiers, and software layers that simulate actual work systems. This setup helps models improve their performance in realistic settings.
Acquisition of Deeptune - recreating enterprise workflows for training environments
In 2026, Mercor acquired Deeptune, a company specializing in reinforcement-learning environments. Deeptune has spent two years recreating hundreds of enterprise applications. These include spreadsheets, customer relationship management (CRM) systems like Salesforce, and other business tools.
Deeptune’s environments allow AI agents to practice professional workflows. Each environment has three parts: the software where work happens, tasks that define what the agent must do, and verifiers that check if the task was completed successfully. These components translate human judgment into signals that guide learning.
Mercor’s goal with Deeptune is to turn these environments into training grounds for AI models. The environments help models learn complex workflows in a safe, controlled setting. This improves their ability to perform in real business environments. It also speeds up the training process by providing realistic practice scenarios.
Model performance measurement - APEX system and real-world tasks
Mercor uses a system called APEX to measure how well models perform. APEX evaluates models on tasks that are economically valuable. It tests models across thousands of real-world workflows to see how they handle different types of work.
The system assesses both the accuracy of the models and their costs. It provides a clear picture of how models perform in practical settings. This helps companies decide if a model is ready for deployment or needs more training. It also guides improvements by highlighting specific areas where models fall short.
Mercor’s focus on real-world tasks ensures that evaluations are relevant. Instead of relying on generic benchmarks, the company tests models on actual workflows. This approach provides more meaningful insights into how models will perform in business environments.
Why it matters - impact on AI development and economic growth
This investment in AI training infrastructure could have a significant impact on the economy. Mercor’s approach aims to make AI models more capable and cost-effective. Better models can perform complex tasks more reliably in real-world settings.
The company’s emphasis on evaluation and post-training reduces the resources needed to improve AI. This can lower costs for businesses adopting AI solutions. It also speeds up the deployment of AI in various industries, from finance to healthcare.
The CEO’s comments suggest that if the ratio of inference spending to employee salaries holds, AI could contribute to faster economic growth. The idea is that more efficient AI models will make workers more productive, leading to higher gross domestic product (GDP). This could result in nearly triple the usual growth rate over five years.
Overall, Mercor’s work supports a future where AI becomes more integrated into everyday business and economic activities. It helps bridge the gap between research and practical application, making AI more accessible and useful.
How OpenSmartRoute helps - what a team routing through OpenSmartRoute does with Mercor raises $750 million at $20 billion valuation to expand AI training
Teams using OpenSmartRoute can benefit from Mercor’s growth and innovations. OpenSmartRoute is an open-source router for AI requests. It directs each request to the best-fit model, agent, or tool from a catalog defined by the team.
OpenSmartRoute scores every candidate based on quality, cost, speed, and safety. The team sets the importance of each factor for every request. Hard rules ensure sensitive data stays on local models, and regional or cost limits are respected. The system learns from outcomes, giving more traffic to models that perform well.
With Mercor’s focus on evaluation and post-training, teams can route requests to models that are proven to perform reliably in real workflows. They can also compare costs and performance across models, optimizing for efficiency. Setting up OpenSmartRoute to use Mercor’s models and evaluation tools helps ensure requests are handled by the best available options, reducing costs and improving results.
By integrating Mercor’s environment and evaluation capabilities, teams can better track model performance and costs. They can also use the 'osr eval' tool to measure routing accuracy on their own prompts. This setup supports continuous improvement and cost savings in AI request handling.
