Blog
Articles, guides, benchmarks, and methodology notes from Mentlio. Every number comes from replayable traces, and we state assumptions where they matter.
Articles and guides
A More Useful Way to Manage AI Coding Costs
How to reduce waste, keep developers moving, and understand the bill without collecting their work.
August 2026 · 7 min read
How to Measure AI ROI in Software Engineering
Attribute AI spend, pick outcome units such as merged PRs or Delivery Points, compute cost per delivered outcome, and remove waste without capping useful usage.
September 2026 · 11 min read
What Is Tokenmaxxing? (And What to Do Instead)
Why maximizing token consumption failed as a proxy for AI productivity by mid-2026, and why token efficiency is the better target for engineering orgs.
September 2026 · 7 min read
Benchmarks and methodology
Terminal-Bench 2.1: The Cost-Quality Frontier Is a Routing Problem
99% of Terminal-Bench 2.1 SOTA accuracy at 25% lower cost.
July 2026 · 8 min read
Routing in the Fable Era: Frontier Quality Without the Frontier Tax
96.32% of Fable route sufficiency at 20.66% lower cost on SWE-Bench Pro.
July 2026 · 8 min read
SWE-Bench Pro Routing Methodology and Results
How Route is scored by replaying router decisions against per-model pass and fail traces, and how savings are computed from model-specific rates.
Spring 2026 · 6 min read
How Mentlio Measures Token-Saver Savings
What each saver measures, how dashboard estimates are computed from pinned benchmark artifacts, and what the numbers deliberately do not claim.
July 2026 · 7 min read