Research Program
The 64GB Frontier
Which open-weight coding models can complete realistic developer tasks on a 64GB M1 Max—and where do they fail?
The 64GB Frontier is an in-progress research program about laptop-scale open AI systems. It begins with practical evaluation of open-weight coding models in a developer-realistic environment on a 64GB M1 Max.
The work is framed around reproducibility, operational constraints, and what actually happens when local models are asked to complete software engineering tasks with tools, tests, and human feedback in the loop.
Focus
- Task completion
- Passing tests
- Latency
- Memory use
- Reliability
- Tool use
- Context handling
- Human intervention
Research Questions
- What classes of developer tasks are practical for open-weight coding models on a 64GB laptop?
- How do latency, memory pressure, and context limits affect real task completion?
- Where does human intervention most improve reliability?
Methodology
- Run developer-realistic tasks against open-weight coding models in a local 64GB M1 Max environment.
- Record task outcomes, tests, latency, memory use, tool behavior, context handling, and intervention points.
Limitations
- Results are not published yet.
- The initial hardware target is intentionally narrow: a 64GB M1 Max laptop.