Artificial Intelligence

Mirendil Raises $1 Billion Series A: The AI Startup Building a System That Can Build Better AI

By mohitdeep1122 • • 7 min read • 📷 3 High-Res Visuals
Share Dispatch
𝕏 Post in Share Y Hacker News
⚡ TL;DR — Executive Summary

The story is not just that a startup raised $1 billion. It is that Mirendil wants to automate the research process behind future AI systems—a potentially important new layer of AI infrastructure.

Mirendil secures $1 billion in Series A funding, bringing its total disclosed capital raised to $1.2 billion as investors back its AI research automation ambitions.
Mirendil secures $1 billion in Series A funding, bringing its total disclosed capital raised to $1.2 billion as investors back its AI research automation ambitions. 🔍 Click to zoom

Mirendil, a San Francisco-based AI research startup founded by former researchers from Anthropic, Google DeepMind, xAI and OpenAI, has raised $1 billion in a Series A round led by Kleiner Perkins and Clearlake Capital. Announced on November 12, 2026, the financing also includes Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32, alongside other new and existing investors. The round follows a $200 million seed financing announced on June 24, meaning Mirendil has secured $1.2 billion in disclosed funding in just 141 days. The latest round is five times the size of its seed. The company has not disclosed its new valuation, although Bloomberg reported in September that it was discussing a potential financing at a $5 billion valuation. That earlier figure should not be treated as the confirmed valuation of this round. Mirendil’s Series A announcement and Bloomberg’s September report provide the funding timeline.

Mirendil is developing AI agents that can write code, run experiments and analyse results, helping researchers automate parts of advanced AI development.
Mirendil is developing AI agents that can write code, run experiments and analyse results, helping researchers automate parts of advanced AI development.🔍 Click image to enlarge

What Mirendil Is Building: AI That Helps Develop AI

Mirendil’s ambition goes beyond building another general-purpose chatbot or coding assistant. It wants to make advanced AI research accessible to smaller teams by creating systems that can handle much of the practical work involved in developing models: writing code, preparing experiments, analysing results, finding errors and improving training processes. Its founding team includes CEO Behnam Neyshabur, who worked at Google DeepMind and Anthropic, and CTO Harsh Mehta, who helped develop Anthropic’s internal automated AI research platform. The wider founding group also includes Shayan Salehian, formerly of xAI, and Tara Rezaei, who has an OpenAI research background. The company says its founding team includes around 20 researchers and engineers from leading AI institutions. The underlying idea is straightforward but ambitious: instead of requiring a large specialist team for every stage of AI development, a smaller group of experienced researchers could direct hundreds of AI agents to carry out much of the work. The company’s founding announcement and Andreessen Horowitz’s investment thesis explain the strategy.

Mirendil-Base and Fern: The First Evidence of Its Approach

Mirendil has started demonstrating this approach through its own model development. In an October 2026 research update, the company said one researcher working with AI agents built Mirendil-Base, a coding-focused model, from scratch in three and a half months using less than $10 million in computing resources. According to the company, agents wrote more than 99% of the code, ran experiments and diagnosed failures, while the researcher made the important decisions about what to test and which results to trust. Mirendil then used this base model to develop Fern, which supports a context window of up to 10 million tokens. In simple terms, a very large context window allows a model to examine much larger codebases, research records and experiment histories without losing track of earlier information. Mirendil’s published evaluations show promising results on selected long-context tasks, but its own benchmark results should be viewed as early technical evidence rather than proof that Fern outperforms established frontier models across the board. The distinction matters: building a model efficiently is an important achievement, but broad performance, reliability and independent evaluation will determine its commercial value. Mirendil’s account of its pretraining experiment and Fern’s technical introduction provide the details.

Lor: The Product That Could Turn the Research Into a Business

The commercial centre of Mirendil’s strategy is Lor, its AI research platform, which was in private beta at the time of the funding announcement. The company describes Lor as a system that can coordinate hundreds of agents working on code and experiments for extended periods. Rather than simply answering a question or generating a short piece of code, these agents are intended to work through longer tasks: run an experiment, inspect the output, identify a problem, change the code and continue testing. Mirendil says it already uses Lor internally for its own research. If the platform works reliably outside the company’s own environment, it could give universities, AI teams, engineering companies and specialist research groups access to capabilities that previously required substantial in-house infrastructure. That is the larger commercial opportunity: Mirendil could become a platform for organisations that want to develop or improve their own AI systems instead of depending entirely on models and research tools controlled by a handful of large AI laboratories. However, Lor’s private-beta status means its wider adoption, pricing, customer demand and commercial performance have not yet been established publicly.

Through Lor and Fern, Mirendil aims to turn AI research automation into a scalable platform for developing, testing and improving advanced AI systems.
Through Lor and Fern, Mirendil aims to turn AI research automation into a scalable platform for developing, testing and improving advanced AI systems.🔍 Click image to enlarge

Why Investors Are Putting So Much Capital Behind Mirendil

The investor interest reflects a wider shift in the AI industry. The first wave of AI investment concentrated on building larger models, securing computing capacity and creating applications around those models. The next opportunity may involve automating the research and engineering process that produces them. If AI agents can reliably handle more of the work involved in training, evaluating and improving models, smaller teams could attempt projects that once required large research departments. Mirendil is also building relationships around the infrastructure needed to pursue this strategy. TechCrunch reported in August that the company had signed a multiyear Google Cloud agreement worth more than $100 million, according to CEO Behnam Neyshabur, to secure computing capacity for its research. Google’s participation in the latest funding round is therefore notable, although its investment and its cloud infrastructure relationship are separate parts of the story. The combination of venture capital, strategic investors and access to computing resources gives Mirendil room to pursue an expensive technical ambition. It does not, by itself, prove that the company has solved the economics of automated AI research. TechCrunch’s report on the Google Cloud agreement provides additional context.

The Opportunity and the Risks Behind the AI Research Automation Bet

Mirendil’s biggest opportunity is to make advanced AI development available to a much wider group of researchers and organisations. A robotics company could use specialised agents to develop models for its machines; a scientific laboratory could build tools around its own research data; and an engineering team could automate parts of model training and testing. But the same ambition comes with difficult questions. AI agents can write code and run experiments, yet researchers still need to verify whether the results are correct, whether the experiments are meaningful and whether a model has learned something useful rather than simply performed well on a narrow test. More autonomous AI research also raises safety and oversight concerns, particularly as systems become involved in improving the models and tools used to develop future systems. Mirendil’s own research describes this as an emerging feedback loop, while acknowledging that its model was built with substantial human direction and that fully autonomous recursive self-improvement has not been achieved. The company must also compete with the internal research platforms of major AI laboratories, which have significant computing resources, talent and access to advanced models. Its opportunity is to make these capabilities available independently; its challenge is to make them dependable, affordable and useful for customers beyond its own research team.

What Comes Next for Mirendil

The $1 billion Series A gives Mirendil substantial resources to develop its models, expand Lor and build the computing infrastructure required for long-running agent workflows. But the next stage should be measured by evidence rather than funding size. Investors and customers will want to see whether Lor moves from private beta into wider use, whether independent researchers can reproduce the productivity gains Mirendil describes, whether Fern and future models perform well on external evaluations, and whether the platform can support scientific and engineering work beyond AI model development. The company has raised $1.2 billion in disclosed funding across its seed and Series A rounds, but it has not announced a confirmed valuation for the latest financing or disclosed enough commercial information to establish its long-term business economics. Mirendil’s story is therefore bigger than a record-sized early funding round. It is a bet that the process of building AI can itself become increasingly automated. If that works, the company could occupy an important position in the next generation of AI infrastructure. If it does not, it will still face the same constraints as other frontier AI businesses: expensive computing, difficult research problems, competition for talent and the need to turn technical progress into a sustainable product.

Sources and Further Reading

📷 Complete Article Visual Gallery (3 Pictures) Click any photo for full-screen view
Mirendil secures $1 billion in Series A funding, bringing its total disclosed capital raised to $1.2 billion as investors back its AI research automation ambitions.
Mirendil secures $1 billion in Series A funding, bringing its total disclosed capital raised to $1.2 billion as investors back its AI research automation ambitions.
Mirendil is developing AI agents that can write code, run experiments and analyse results, helping researchers automate parts of advanced AI development.
Mirendil is developing AI agents that can write code, run experiments and analyse results, helping researchers automate parts of advanced AI development.
Through Lor and Fern, Mirendil aims to turn AI research automation into a scalable platform for developing, testing and improving advanced AI systems.
Through Lor and Fern, Mirendil aims to turn AI research automation into a scalable platform for developing, testing and improving advanced AI systems.
mohitdeep1122 • TechFront360 Editorial Desk

Published by TechFront360.com following our strict fact-checking, benchmark verification, and independent technology journalism standards.

More in Artificial Intelligence