‘Bringing the cost of replicating strong coding agents down to a few hundred dollars will unlock research that simply wasn’t possible before’, AI2 said.
The non-profit Allen Institute for AI (AI2) has launched a family of open-source coding models targeting independent developers and SMEs.
The idea behind the launch is to simplify building coding agents for any code base, with the added benefit of cost effectiveness, the company explained.
The first release from the Open Coding Agents suite is called SERA (Soft-verified Efficient Repository Agents), which is available in two versions.
SERA-32B is a 32bn-parameter model that can solve 54.2pc of problems classed as ‘SWE-bench verified’, part of a benchmark used for evaluating large language models (LLMs) on real-world software engineering tasks. This success rate beats previous open-source models of comparable sizes such as Qwen3-Coder, and closed models such as Mistral3’s Devstral Small 2.
SERA-8B is an 8bn-parameter model that solves 29.4pc SWE-bench verified problems.
AI2 collaborated with Nvidia to optimise SERA’s interface in order to help researchers and developers get the most out of the new models in production environments, the company said in a launch blog post.
Every component of the family is open, including the models, training recipes and integration with Anthropic’s Claude Code, the company said.
SERA can be launched with a single line of code, making it easy to use even for those without LLM experience, it added. The company has also released training data for researchers to inspect before recommending tweaks.
According to AI2, the total cost to use SERA to reproduce the performance levels of the best existing open-source result is around $400, which is around 25 times cheaper than many existing approaches, while the total cost to reproduce top open-weight models in industry is around $12,000.
“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,” the company said in the post.
AI2 was established in 2014 by the late co-founder of Microsoft, Paul Allen. The institute aims to deliver “real-world impact” via large-scale open models, data and robotics.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.