Shift-1 router cuts coding agent spend up to 55% while keeping SWE-bench accuracy
ypatil125 · x · 2026-10-27
Applied Compute released Shift-1, a frontier-quality model router trained for coding, and shared its research. Routing between GPT-6 Astra and GPT-6.1 Sol, it cuts Codex spend by 55% while preserving SWE-bench Verified accuracy and user satisfaction; in Claude Code it routes between Opus 5.5 and Sonnet 5.5, cutting spend 24% with a small accuracy hit.
Key design choices:
- Drop-in replacement endpoint: works without controlling the harness, complementing subagent-based cost reduction rather than replacing it.
- Per-request routing: difficulty is hard to assess upfront, so Shift-1 can update its routing decision mid-task instead of assigning per task.
- Cost-aware decisions and outcome-grounded router training to balance quality and cost.
Yash Patil praised the work, noting routing policies differ by domain and model family, supporting the thesis that intelligence is "jagged" — there's a best model for each job.
More from coding & agent
- BELL: Open-Source Stack Combines Browserbase, Eve, iMessage and Stripe Link for Personal Agents — jeff_weinstein · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Two engineers rebuilt a FedEx supplier's billing core in 3.5 months vs an estimated 7-8 — alex_verem · 2026-10-27
- GPT-6.1 Terra scores 48 on AA Coding Agent Index at one-third the cost of GPT-5.6 Terra — ArtificialAnlys · 2026-10-27
- Non-coder vibe-codes a Godot JRPG with Claude Code in two weeks, music and art synced — ADHDavid · 2026-10-27
- Paradev (fka Superset) raises $11.5M seed led by USV to orchestrate armies of coding agents — rebeccakaden · 2026-10-27