How we make recommendations
Recommendations are editorial judgement, made transparent and kept current.
Neutral strength tiers
We map provider models onto four neutral strengths — Light, Standard, Strong, Maximum — through versioned data. We do not claim a model from one vendor is exactly equivalent to another; the tiers describe the kind of work a model is suited to.
What every recommendation carries
- A recommended platform, model, and strength.
- Best-value, lowest-cost, and premium alternatives.
- An estimated token and cost range, and expected duration.
- The reasoning — why this strength, when to upgrade, when cheaper is fine.
- A confidence level, methodology version, and last-reviewed date.
Why versioning matters
Models and prices change constantly. Each recommendation is stamped with the methodology version and the date it was last reviewed, so advice that has gone stale is visible rather than silently wrong.
What we won’t do
- Declare a single model “the best” without evidence.
- Hide the assumptions behind a budget range.
- Present editorial judgement as measured fact.