Microsoft's Echoverse: Deep Synthetic Environments Boost Computer-Use Agents to 65% Accuracy
omarsar0 · x · 2026-08-01
Microsoft has published new research named Echoverse, focusing on training computer-use agents at scale.
- Approach: While recent pipelines generate synthetic environments in bulk, environment quality becomes the bottleneck. Echoverse compiles specifications into stateful applications, grades tasks against the application's own database, and runs a co-evolution loop. This loop reads every graded rollout twice: once as repairs to the environment and its verifier, and once as a training signal for the model.
- Results: On the same domains, shallow environments pushed live-site accuracy below the base model (from 80.0% down to 75.0%). Deep environments raised it significantly (from 80.0% to 85.0% and 48.0% to 65.0%).
- Impact: Repairing a single environment lifted the trained model's accuracy from 16.2% to 38.5%. Across twelve environments, a 9B model improved from 36.5% to 67.1% on fourteen evaluation splits, performing within fourteen points of much larger models.
More from coding & agent
- 'LLM code is sloppy' rebutted with real ffmpeg code: identical if/else branches in a top library — kingslyj · 2026-10-16
- Office-editing tool calls need file, worktree and unit: lessons from dsh-univer-office — FunDistribution6537 · 2026-10-16
- Architect MD workflow: Gemini one-shot outputs that beat Opus, via 5-lens plan critique — oikk01 · 2026-10-16
- Prove a coding agent actually ran the tests: check artifacts, not approvals — ComprehensiveShake76 · 2026-10-16
- Building an acceptance harness for AI spreadsheet edits: allowlists, recalc checks, idempotency — ukdicha-modakkk · 2026-10-16
- Agent actions need checks outside the model: NVIDIA demos lean on guardrails amid wave of real agent incidents — Creamy-And-Crowded · 2026-10-16