← The Wire

Research2026-06-14 · source-backed

DeNovoSWE lifts whole-repo generation from 5.8% to 47.2%.

Confidence
source-backed
Sources
1
Redaction
redacted

Story

The dataset has 4,818 instances for generating entire repositories from documentation, and fine-tuning Qwen3-30B-A3B on it jumps BeyondSWE-Doc2Repo performance from 5.8% to 47.2% (arXiv). That's a large move on a long-horizon doc-to-repo task, real signal toward agents that scaffold whole codebases instead of patching files. Still far from "describe an app, get a working repo," but the trajectory on this specific axis is steep.

Related stories

Each link below shares sources, entities, or timing with this story.

  1. Shared entities / Earlier coverage / Tension

    Nemotron Cascade 2: NVIDIA's 30B-A3B MoE Outperforms Qwen3.5-35B

    Both cover A3B, Qwen3; earlier A3B coverage from 2026-03-22; pushes against this story (vs).

  2. Shared entity: Still / Same source domain / Earlier coverage / Tension

    RL-trained drone racers beat a national champion, trained in 27 hours on one RTX 4090.

    Both cover Still; reported by the same outlet (arxiv.org); earlier Still coverage from 2026-06-08.

  3. Shared entity: Qwen3 / Same source domain / Earlier coverage / Tension

    RefineRL: 4B Models Match 235B via Self-Refinement RL.

    Both cover Qwen3; reported by the same outlet (arxiv.org); earlier Qwen3 coverage from 2026-04-21.

  4. Shared entities / Earlier coverage

    llama.cpp Becomes an Agentic Platform: MCP Client + Autoparser Merged Same Day

    Both cover A3B, Qwen3; earlier A3B coverage from 2026-03-06.

  5. Shared entity: Qwen3 / Same source domain / Earlier coverage

    Research Papers

    Both cover Qwen3; reported by the same outlet (arxiv.org); earlier Qwen3 coverage from 2026-02-25.

  6. Shared entity: Qwen3 / What happened next / Tension

    Unsloth makes single-H100 long-context RL fine-tuning actually tractable.

    Both cover Qwen3; picks up the Qwen3 thread on 2026-06-20; pushes against this story (versus).

  7. Same source domain / Shared topic / Tension

    Auto-mined agent skills are interpretable but barely move the needle.

    Reported by the same outlet (arxiv.org); overlapping topics (agent, generating); pushes against this story (against).

  8. Your AI-Generated Code Compiles. It Doesn't Work.

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

Source trail

Entities

Provenance

AI generated
no
Story unit
2026-06-14-denovoswe-lifts-whole-repo-generation-from-5-8-to-47-2
Labels
source-backed, canonical briefing excerpt