Hexo Labs releases SIA, an open-source framework for recursive AI agent self-improvement through weight and workflow updates
It scored 0.701 on LawBench, beating Claude Code's 0.173.
#417Xin Eric Wang ✈️ CVPR 2026@XWANG_LKKunal and Vignesh are among the rare founders who are genuinely driven by a deep passion for research and a long-term vision for the future.
Read the paper: https://arxiv.org/abs/2605.27276
Github Repo: https://github.com/hexo-ai/sia
Rohan Paul@rohanpaul_aiBig release - Open Source Recursive Self Improvement from @hexoai Shows AI agent can improve both how it works and what it internally knows after seeing its own task results. i.e. by repeatedly training on its own task feedback, not by relying on a human to hand-code every strategy. Most agents today are frozen workers: you can give them better prompts, better tools, better retry rules, and better code, but the actual model usually stays the same. SIA (Self Improving AI framework) changes the outer workflow, called the harness, and also changes the model’s weights, which are the internal settings that store learned patterns. which means task feedback changes the model’s internal parameters, pushing it toward domain knowledge. The paper reports a 56.6% gain on LawBench, 91.9% runtime reduction on GPU kernels, and 502% improvement on single-cell RNA denoising over baseline.
Rohan Paul@rohanpaul_aiBig release - Open Source Recursive Self Improvement from @hexoai Shows AI agent can improve both how it works and what it internally knows after seeing its own task results. i.e. by repeatedly training on its own task feedback, not by relying on a human to hand-code every strategy. Most agents today are frozen workers: you can give them better prompts, better tools, better retry rules, and better code, but the actual model usually stays the same. SIA (Self Improving AI framework) changes the outer workflow, called the harness, and also changes the model’s weights, which are the internal settings that store learned patterns. which means task feedback changes the model’s internal parameters, pushing it toward domain knowledge. The paper reports a 56.6% gain on LawBench, 91.9% runtime reduction on GPU kernels, and 502% improvement on single-cell RNA denoising over baseline.
Rohan Paul@rohanpaul_aiBig release - Open Source Recursive Self Improvement from @hexoai Shows AI agent can improve both how it works and what it internally knows after seeing its own task results. i.e. by repeatedly training on its own task feedback, not by relying on a human to hand-code every strategy. Most agents today are frozen workers: you can give them better prompts, better tools, better retry rules, and better code, but the actual model usually stays the same. SIA (Self Improving AI framework) changes the outer workflow, called the harness, and also changes the model’s weights, which are the internal settings that store learned patterns. which means task feedback changes the model’s internal parameters, pushing it toward domain knowledge. The paper reports a 56.6% gain on LawBench, 91.9% runtime reduction on GPU kernels, and 502% improvement on single-cell RNA denoising over baseline.
Rohan Paul@rohanpaul_aiBig release - Open Source Recursive Self Improvement from @hexoai Shows AI agent can improve both how it works and what it internally knows after seeing its own task results. i.e. by repeatedly training on its own task feedback, not by relying on a human to hand-code every strategy. Most agents today are frozen workers: you can give them better prompts, better tools, better retry rules, and better code, but the actual model usually stays the same. SIA (Self Improving AI framework) changes the outer workflow, called the harness, and also changes the model’s weights, which are the internal settings that store learned patterns. which means task feedback changes the model’s internal parameters, pushing it toward domain knowledge. The paper reports a 56.6% gain on LawBench, 91.9% runtime reduction on GPU kernels, and 502% improvement on single-cell RNA denoising over baseline.
The big deal is that its Improvement and evolution on loop.
Task attempt, feedback, scaffold change, model update, better attempt, more feedback.
So if agents can repeatedly convert experience into both better behavior and better internal knowledge, then human engineering stops being the only path by which AI systems improve.
Rohan Paul@rohanpaul_aiBig release - Open Source Recursive Self Improvement from @hexoai Shows AI agent can improve both how it works and what it internally knows after seeing its own task results. i.e. by repeatedly training on its own task feedback, not by relying on a human to hand-code every strategy. Most agents today are frozen workers: you can give them better prompts, better tools, better retry rules, and better code, but the actual model usually stays the same. SIA (Self Improving AI framework) changes the outer workflow, called the harness, and also changes the model’s weights, which are the internal settings that store learned patterns. which means task feedback changes the model’s internal parameters, pushing it toward domain knowledge. The paper reports a 56.6% gain on LawBench, 91.9% runtime reduction on GPU kernels, and 502% improvement on single-cell RNA denoising over baseline.