OpenDev: Terminal-Native Coding Agent Academic Paper

Shared entity: Terminal / Same source domain / Shared topic / What happened next / Tension

The best frontier agent clears 15% of real terminal tasks. Read that again.

Both cover Terminal; reported by the same outlet (arxiv.org); overlapping topics (agent, model, paper).

Shared entity: Terminal / Same source domain / Shared topic / What happened next

CompactionRL makes context compaction a trained policy, +7 on SWE-bench.

Both cover Terminal; reported by the same outlet (arxiv.org); overlapping topics (agent, coding, compaction, context).

Ouroboros has been live-evolving in public for 161 days and claims SOTA on three benchmarks.

Both cover Terminal; reported by the same outlet (arxiv.org); overlapping topics (agent, context, paper).

A second agent decides when to remind the first one

Both cover Terminal; reported by the same outlet (arxiv.org); overlapping topics (agent, context, memory).

Same source domain / Shared topic / Tension

Partition Your Agent Memory Into Heads: Retention Goes From Under 30% to 74% at 896K Tokens

Reported by the same outlet (arxiv.org); overlapping topics (agent, compaction, context, memory, paper); pushes against this story (but).

Shared entity: Terminal / Shared topic / What happened next / Tension

Cohere open-sourced a 30B coding model that runs on a single H100. Agentic coding without the frontier bill.

Both cover Terminal; overlapping topics (agent, coding, model); picks up the Terminal thread on 2026-06-10.

Cline's Open-Source SDK Outscores Claude Code on the Same Model. The Agent Lock-In Myth Is Over.

Both cover Terminal; overlapping topics (agent, context, model); picks up the Terminal thread on 2026-05-14.

Kimi K2.6: A 1-Trillion-Parameter Open-Weights Model Just Beat GPT-5.4 on SWE-Bench Pro

Both cover Terminal; overlapping topics (agent, coding, model); picks up the Terminal thread on 2026-04-21.