Home / Methodology
How we test
The point of EditBench is that every tool gets exactly the same input and is judged by exactly the same rules, and that you can see both.
1. The test video
- One 30–60 minute royalty-free video, the same file for every tool in a round.
- Chosen to look like real creator footage: talking heads, more than one speaker, some on-screen content.
- We publish the source and licence with each round so anyone can repeat the test.
- We never use client or private footage.
2. Plans and settings
- Round 1 uses each tool's free plan or free trial. That's what most people try first, and it keeps us from being dependent on any vendor.
- Free plans often add a watermark or limit resolution. We disclose this on every tool page and don't penalise the watermark itself in the scores.
- We use each tool's default clipping settings unless a setting is required to make it work (for example choosing 9:16). Any changed setting is listed with the results.
- Where a feature isn't on the free plan, we say so rather than guess. Submagic's clipping is a paid add-on, so it's tested for captions only in round 1.
3. The scoring rubric
Each tool's clips are scored on five criteria, each from 1 (poor) to 10 (excellent). Scores are per tool, based on all the clips it produced from the test video.
Idea completeness · 30% of overall
Does each clip contain one complete thought that makes sense to someone who hasn't seen the full video? It needs a setup, a point and a payoff, without needing outside context.
- 9–10
- Almost every clip stands alone: clear hook, complete point, satisfying end.
- 5–6
- About half the clips stand alone; the rest need context or stop mid-idea.
- 1–2
- Clips are mostly fragments that don't make sense on their own.
Cut points · 25% of overall
Are the start and end points clean? We check for clips that start or end mid-word or mid-sentence, dead air at either end, and awkward breaths or filler at the edges.
- 9–10
- Cuts land on natural sentence boundaries; no trimming needed.
- 5–6
- Several clips need a manual trim at the start or end.
- 1–2
- Most clips cut into or out of words.
Framing · 20% of overall
When the tool reframes horizontal footage to vertical, does it keep the active speaker (and anything important on screen) in frame, and does it switch smoothly between speakers?
- 9–10
- Speaker always well framed, smooth switches, nothing important cropped.
- 5–6
- Occasional wrong speaker, late switches or cropped faces.
- 1–2
- Frequently frames the wrong person or empty space.
Caption accuracy · 15% of overall
How accurate are the auto-captions against our hand-checked transcript? We count wrong, missing and extra words, plus timing that drifts out of sync.
- 9–10
- Near word-perfect, in sync, names and jargon right.
- 5–6
- Noticeable errors in most clips; would need a correction pass.
- 1–2
- Errors in nearly every line.
Cost per usable clip · 10% of overall
The plan price divided by the number of usable clips the tool produced from the test video. A clip is usable if we'd post it after no more than a quick trim. We publish the raw cost and convert it to a 1–10 value score on a fixed scale published with each round.
- 9–10
- Lots of usable clips for little or no money.
- 5–6
- Average cost per usable clip across the tools tested.
- 1–2
- Few usable clips for the price.
Overall score
Overall = 0.30 × Ideas + 0.25 × Cuts + 0.20 × Framing + 0.15 × Captions + 0.10 × Value, rounded to one decimal place.
A usable clip is one we would post after no more than a quick trim: it makes sense on its own, the cuts are acceptable, the speaker is in frame and the captions need at most one or two fixes.
4. Monthly re-tests
AI tools update constantly. We re-run the test every month, publish the new scores, and note what changed and why. Older rounds stay available so you can see the history.
5. Independence
EditBench is run by The EditBench team and isn't owned by, or affiliated with, any tool we test. No company can pay for a score, see results before they're published, or ask us to change them. Tools are tested on the same terms whether or not we have any commercial relationship with them.
6. Affiliate disclosure
Some links on EditBench may be affiliate links. If you sign up for a paid plan through one, we may earn a commission at no extra cost to you. Commission never affects scores, rankings or what we write. A tool with no affiliate programme is tested and ranked exactly the same way.
If we ever run sponsored placements, they will always be clearly labelled Sponsored and kept separate from the leaderboard.
7. Corrections
If a tool changes its free plan or we get something wrong, we fix it and note the date. Free-plan facts on this site were last checked on 22 October 2026.