Research Papers Topics
Models and systems that synthesize, complete, or reason about programs.
Code generation is where language models meet executable reality. A useful coding model cannot only produce fluent text; it has to respect syntax, project context, hidden tests, security constraints, and the long-term cost of maintenance.
The strongest research pattern is generation plus verification. AlphaCode showed that sampling many programs and filtering them can solve contest problems better than trusting one answer. Code Llama made open code-specialized models practical for local deployment and fine-tuning. The next step for this topic is coding agents that read repositories, run tests, and revise their own patches.
Code Generation · Google DeepMind
DeepMind's AlphaCode averaged a top 54.3% ranking on Codeforces contests with 5,000+ participants by generating up to a million candidate programs per problem, then filtering and clustering them down to ten submissions.
Code Llama continues training Llama 2 on code, reaching up to 67% on HumanEval and 65% on MBPP, the best open scores at its release, with infilling, instruction following, and 100k-token context support.
Code Generation · Google DeepMind
DeepMind's AlphaCode averaged a top 54.3% ranking on Codeforces contests with 5,000+ participants by generating up to a million candidate programs per problem, then filtering and clustering them down to ten submissions.
Code Llama continues training Llama 2 on code, reaching up to 67% on HumanEval and 65% on MBPP, the best open scores at its release, with infilling, instruction following, and 100k-token context support.
AI Agents · University of Illinois Urbana-Champaign
This survey reframes code not as a thing agents generate but as the executable substrate they run on, mapping 40-plus systems across three layers — interface, mechanisms, multi-agent scaling — plus seven open problems.
SkillsVote treats agent skills as a governed library — profiling a million-scale corpus, recommending skills before a run, and gating updates after. Offline evolution lifts GPT-5.2 on Terminal-Bench 2.0 by up to 7.9 pp.
AI Agents · TokenRhythm Technologies
Claw-SWE-Bench evaluates OpenClaw-style coding-agent harnesses on 350 GitHub issue tasks. OpenClaw jumps from 19.1% to 73.4% Pass@1 with a full adapter.
AI Agents · Shanghai Jiao Tong University
SWE-Explore isolates the repo-exploration stage of coding agents over 848 issues. Agentic explorers crush BM25 (HitFile 0.65 vs 0.08), but line-level recall stalls at 0.15-0.20, and that gap is what limits repairs.
Code Generation · Google DeepMind
DeepMind's AlphaCode averaged a top 54.3% ranking on Codeforces contests with 5,000+ participants by generating up to a million candidate programs per problem, then filtering and clustering them down to ten submissions.
AI Agents · University of Illinois Urbana-Champaign
This survey reframes code not as a thing agents generate but as the executable substrate they run on, mapping 40-plus systems across three layers — interface, mechanisms, multi-agent scaling — plus seven open problems.
Code Generation · University of Waterloo
Code2LoRA trains a hypernetwork to emit a repo-specific LoRA adapter for a code model with no inference-time token cost — 66.2% in-repo and 63.8% cross-repo exact match, plus an Evo variant that tracks diffs with a GRU.
Code Llama continues training Llama 2 on code, reaching up to 67% on HumanEval and 65% on MBPP, the best open scores at its release, with infilling, instruction following, and 100k-token context support.
AI Agents · TokenRhythm Technologies
Claw-SWE-Bench evaluates OpenClaw-style coding-agent harnesses on 350 GitHub issue tasks. OpenClaw jumps from 19.1% to 73.4% Pass@1 with a full adapter.
AI Agents · Shanghai Jiao Tong University
SWE-Explore isolates the repo-exploration stage of coding agents over 848 issues. Agentic explorers crush BM25 (HitFile 0.65 vs 0.08), but line-level recall stalls at 0.15-0.20, and that gap is what limits repairs.
Code Generation · Google DeepMind
DeepMind's AlphaCode averaged a top 54.3% ranking on Codeforces contests with 5,000+ participants by generating up to a million candidate programs per problem, then filtering and clustering them down to ten submissions.
AI Agents · University of Illinois Urbana-Champaign
This survey reframes code not as a thing agents generate but as the executable substrate they run on, mapping 40-plus systems across three layers — interface, mechanisms, multi-agent scaling — plus seven open problems.
Code Generation · University of Waterloo
Code2LoRA trains a hypernetwork to emit a repo-specific LoRA adapter for a code model with no inference-time token cost — 66.2% in-repo and 63.8% cross-repo exact match, plus an Evo variant that tracks diffs with a GRU.
Code Llama continues training Llama 2 on code, reaching up to 67% on HumanEval and 65% on MBPP, the best open scores at its release, with infilling, instruction following, and 100k-token context support.
Mellum 2 is JetBrains' open-weight 12B Mixture-of-Experts code model that activates only 2.5B parameters per token, matching dense 4B-14B baselines on software tasks at a fraction of the per-token compute.
SkillsVote treats agent skills as a governed library — profiling a million-scale corpus, recommending skills before a run, and gating updates after. Offline evolution lifts GPT-5.2 on Terminal-Bench 2.0 by up to 7.9 pp.