Skip to content
View original post on X: Sophia Yang
Published
Pick62/100

Mirendil announces Mirendil-Base, a coding model pretrained in 3.5 months for under $10M

AIMirendil, a lab working on recursive self-improvement, announced its first model, Mirendil-Base. One researcher pretrained it from scratch in 3.5 months with less than $10M in compute, and agents wrote over 99% of the code.

AIWhy it matters.

The post gives a concrete cost, timeline, and held-out code comparison, which helps readers gauge how far one researcher's pretraining effort can go.

Sophia YangVerified on X
@sophiamyang
View on X

TL;DR:

- Mirendil, anothet lab working toward recursive self-improvement (AI that builds AI), announced its first model, Mirendil-Base

- One researcher and a team of agents pretrained it from scratch in 3.5 months, for under $10M all-in

- It isn't released yet: no weights or API, and post-training is still underway

- Agents wrote 99%+ of the code; training used only public data, with no distillation and no synthetic data

- It beats GLM-5.3 and DeepSeek-V4-Pro-Base at predicting held-out code (0.280 vs 0.293 and 0.301 bits per byte, lower is better)

- Kimi-K3 still leads (0.193 vs 0.222 on 7 outside repos), but Mirendil's Large and XL versions, still mid-training, already pass it on Mirendil's own code

- Not shown yet: model size and task benchmarks like SWE-bench

Behnam Neyshabur@bneyshabur
We formally announced Mirendil less than 4 months ago. Today, I'm excited to share an important milestone for: Pretraining at scale without a pretraining team. Mirendil's mission is to solve Recursive Self-Improvement (RSI) and use it to democratize frontier AI development, ultimately accelerating science and technology. At Mirendil, when we think about RSI, we think holistically about the company as an intelligent entity that self-improves toward its goal. From that perspective, the most powerful notion of acceleration is reducing the time and resources (number of human experts, compute, and other costs) needed to achieve ambitious goals. Given that pretraining is perhaps the most important yet most challenging part of frontier AI development, we decided to run an ambitious controlled experiment: Starting fully from scratch, can we significantly reduce the time, number of human experts, and compute resources needed to go all the way to successfully pretraining a strong coding model at scale? Mirendil-Base is the result of this experiment. Our first pretrained model was built by one researcher in 3.5 months, with less than $10M in compute. This includes all the research and small-scale experiments it took to get there, as well as the final training run. The recipe holds up as the models get bigger. For reference, Mirendil-Base is in the 100B-500B parameter range, Mirendil-Base-Large in the 500B-2T range, and Mirendil-Base-XL in the 2T-10T range. We're beginning to see a world where a single researcher, empowered by AI, can take on work that once required an entire research organization. As these systems improve, the bottleneck shifts from the size of a team to the ambition and judgment of the people directing them. This has profound implications for who gets to build AI and, ultimately, who gets to push the frontiers of science and technology forward. We’ll share independent evaluations, post-training benchmarks and API-access sooner than you think so stay tuned!
View quoted post on X
Source details

Sophia Yang · x.comPublished · added here

Original title: TL;DR:

Topics

AI agentsAI codingData & trainingModel releasesEvals & benchmarks