Once you learn about this, you start seeing it everywhere.

gpt-6.2-astra is rolling out! You can run gpt-6.2-astra straight from Codex CLI. In the terminal: `codex --model gpt-6.2-astra`

Our opening animation and blog didn't fully show the interesting dynamics for languages like CJK that have multi-byte characters. In Chinese most characters take up 3 bytes, and when learning end-to-end from bytes it's not even trivial to get the model to chunk these into characters. Early in training this H-Net incorrectly sub-divides characters, but it eventually figures it out. Actually this was a prominent issue in the original H-Net - and even BPE doesn't get it right on rare characters. The blog has some examples!

I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data. Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less? Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.

I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data. Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less? Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.

Introducing Fern, Mirendil’s first model: pretrained from scratch, with up to 10 million tokens of context. That’s ~0.5M lines of code or days of an agent’s work - all in context at once. Built for high-volume agentic work in AI R&D: auditing training data, debugging training runs from logs, and reading agent traces end to end. Fern’s next-token prediction keeps improving as context grows to 10M tokens. Fern 0.1 is Mirendil-Base after a short round of post-training. It’s the first and smallest of three models, with two larger ones now being pretrained. We’ll ship frequent updates and make the model available soon.

what I do to my girlfriend during sex

We’re about to learn that most code written over the last 20 years was complete slop and most software developers were never that good

The good news: you can mid-train yourself to the frontier! At a fraction of the cost! Without distillation! What we did: We took Qwen3.5 14B and mid- and post-trained it to beat every other open model and perform as well as the very best frontier models. This is no simple toy task or format fixing, we are training the model to understand and reason over DNA sequences. Bio is the next frontier for AI and universal access will be crucial to make progress. The fun part: because we train the model on both natural language and DNA sequences you can actually 'chat to your DNA': Ask it questions, ponder the impact of mutations, or ask it to generate a small sequence. We'll soon see the frontier labs move away from math and cyber towards bio. It already started, and we believe researchers should have access to capable models. We are releasing the full recipe, so you can do it, too! Model: huggingface.co/HuggingFaceBi… Blog: huggingface.co/spaces/Huggin…

Today, we're sharing a major milestone in our mission to democratize frontier AI development: Pretraining at scale without a pretraining team. We set out to answer an ambitious question: Can we significantly reduce the time, human expertise, and compute needed to pretrain a competitive foundation model from scratch? The result is Mirendil-Base, our first pretrained coding model, built by one researcher in just 3.5 months with less than $10M in compute, including all research, experimentation, and the final training run. Our approach is already showing promising results as we scale to even larger models. This is an important step toward a future where a single researcher, empowered by AI, can accomplish what once required an entire research organization. As these capabilities improve, we believe frontier AI development can become accessible to far more people and teams. More to come soon, including independent evaluations, post-training benchmarks, and API access.

Cancer patients with the same diagnosis often have very different tumor biology. Molecular testing can reveal underlying differences, but that's slow and expensive. Introducing Holon-S1, our platform to predict genes and biomarkers from routine pathology slides. • beats two other models at predicting spatial gene activity in a controlled comparison • has identified potentially novel biomarkers in breast cancer and colon cancer • evaluated across 16 organ types and more than 15 scanners, using slides from both formalin-fixed, paraffin-embedded and fresh-frozen tissue. We believe Holon-S1 has the potential to help researchers find new biomarkers in slides they already have, discover treatment targets, and bring more personalized cancer care within reach!

You can just go to LCR.net and give your agents access to a fleet of 10s of thousands of browsers that have the ability to route requests through millions of real residential IP addresses.

Glad we have some confirmation that the sudden increase in 3D reasoning abilities by frontier models is specifically trained for via RL rather than some emergent capability.

Today we launch NeuroRisk, the first embryo test for autism and schizophrenia that looks for mutations in evolutionarily important genes. These mutations are individually rare but over 10% of people carry at least one mutation in this class.