Route Anything: model routing yields 8.82x quality per dollar, 25.5x on coding tasks
rohanpaul_ai · x · 2026-11-11
A Route Anything study from UC Berkeley, NUS and other labs systematically answers whether model routing pays off:
Scale
- 452 candidate models/policies, 25 benchmarks (from text QA to robot control), 374,401 scored runs
Key findings
- Routing boosts quality per unit cost by 8.82x on average, losing under 0.5 points
- A perfect per-input oracle router would beat the best single model by 5.4 points, but learned routers gained only 0.3 across 16 benchmarks — routing's value is saving money, not getting smarter
- On coding tasks, routing reached 25.5x the dollar efficiency of always calling Opus 4.6 at a 0.44-point loss; text QA sees 7.1x
- Video generation and robot manipulation also show sizable gains
Practical advice: use routers mainly to move work to cheaper models when quality is close; compare candidates on the same inputs before trusting routing gains.
Related event: Route Anything: Model Routing Cuts Costs 8.82x on Average(2 posts)→
More from Infra
- FlashAttention co-creator Dan Fu on kernels, inference economics, and owning vs leasing GPUs — realDanFu · 2026-11-11
- Self-hosting agent sandboxes on AWS costs a quarter of Daytona's list price — KlausCodes · 2026-11-11
- The Jensen Huang Affair Is the Least Interesting Part: Tracing the $4.8B Nvidia-to-Megaspeed GPU Story — imdigitalashish · 2026-11-11
- SoftBank posts $7B quarterly loss after turning $2B Intel stake into $10.5B and hedging it — realsohamparekh · 2026-11-11
- Email API for agents: appending RFC 5322 copies to Sent via IMAP to fix missing outbound mail — kumard3 · 2026-11-11
- Flide: open-source inference engine with exact state forking claims up to 9.5x speedup over vLLM — _akhaliq · 2026-11-11