Ai2 Newsletter
February 2026
Top story - Introducing SERA—Open coding agents you can customize
It's been a big month for open, customizable coding models. We're excited to share Ai2 Open Coding Agents, starting with SERA—a family of open models and an efficient training recipe that makes it practical to specialize a coding agent to any repo, including private codebases.
Closed coding agents don't know your internal APIs, conventions, or tooling. With SERA, you can generate realistic, agent-style training data from a repo and fine-tune quickly—so the agent learns how your codebase actually works. If you're a small to mid-sized business or independent developer working with customer data in ways no public model has ever seen, SERA lets you specialize a model to your stack without exposing that data to an outside provider.
The key insight behind SERA is that high-quality synthetic training data should mirror how a developer works on a problem rather than the precise details of correct code. Combined with our soft-verification approach, this makes it straightforward to generate massive amounts of agentic training data for any codebase—so a repo with thousands of functions can yield tens of thousands of varied trajectories at low cost.
One result we're especially excited about: a smaller, open model can replicate and even exceed the performance of a more capable "teacher" coding agent when specialized to a particular codebase. This means you can compress frontier-level performance into a small, easily deployable model tailored to your private data.
We believe bringing the cost of replicating strong coding agents down to a few hundred dollars will unlock research that simply wasn't possible before. Instead of being limited to a handful of well-funded labs, agentic coding can become a widely accessible practice. Case in point: SERA was built largely by a single Ai2 researcher.
Read the full announcement on our blog, download the models from Hugging Face, or dive into the technical details in our report. Ready to try it? Get started with the SERA CLI on GitHub or install directly from PyPI.
HiRO-ACE
A two-stage AI framework, developed with NOAA, that generates high-resolution (3 km) precipitation data for any region of the globe. HiRO-ACE lets researchers produce decades of climate simulations in a day on a single GPU—work that would take months with traditional methods.
Theorizer
A multi-LLM research prototype that reads scientific literature and outputs structured ⟨LAW, SCOPE, EVIDENCE⟩ theories—helping scientists get oriented in a new domain in minutes rather than months. We also released a dataset of ~3,000 generated theories from AI/NLP research to build on.
New in AstaLabs: Paper+Figure QA
Paper+Figure QA is an experimental new tool in our AstaLabs platform that lets you ask questions about any paper—including figures, tables, and text. Just enter a paper title or Semantic Scholar URL, ask a question, and go. Use it for general reasoning, comparing across multiple figures, or pulling insights from a specific table or chart.
Paper+Figure QA is designed to support scientists with diverse visual needs – from sighted researchers to those who are blind or low-vision – across all scientific domains. It's early and still evolving, so expect some rough edges. We'd love your feedback as we improve it!