- 0:00–0:06
Contract Q&A • Name the job first • Long Arabic contracts • Context length decides
The “Contract Q&A” badge stays fixed in its own lane. Five long contract pages arrive one after another and build a tall stack; a cyan bracket traces the stack’s full length. “Long Arabic contracts” names the pages. A “Context length decides” tag travels to the stack and clips onto it.
- 0:06–0:12
Contract Q&A • Let benchmarks earn a test • Developer's own figures • Developer's benchmark • Shortlist • Your Arabic test
The article’s benchmark figure appears: the developer’s benchmark report, a solid arrow to a shortlist of three blank cards, and a dashed path on to the open Arabic test ring. A marker travels along the solid arrow from “Developer's benchmark” and stops at “Shortlist”, which is outlined. The dashed path to “Your Arabic test” stays unlit and the ring stays empty; a dashed outline settles round it, so the test is still to run. “Developer's own figures” sits under the title.
- 0:12–0:18
Contract Q&A • Check the context fit • Context-limit checkpoint • Recorded as a fit issue
A short window frame appears under the caption “Context-limit checkpoint”, which carries no number. A long contract strip slides in behind the window and stops centred on it, overhanging both ends; the overhanging parts dim and dashed lines mark the window’s limits. A “Recorded as a fit issue” note clips on under the window. No score appears.
- 0:18–0:24
Contract Q&A • Change only the model • Documented chat format • Same instructions • Same retrieved passages • Same output settings
The three locked input cards of the article’s figure stay still: “Same instructions”, “Same retrieved passages” and “Same output settings”. A cyan arrow traces from each card into one model slot. The blank block in the slot swaps once with the dashed outline behind it: the front block moves back and turns dashed while the outline comes forward and turns solid. A “Documented chat format” tag clips onto the slot. No run result is shown.
- 0:24–0:30
Contract Q&A • Re-run when a checkpoint ships • New checkpoint • Same test again • At least quarterly
A candidate lane of three blank tokens leads along a dashed path to the open test ring. A “New checkpoint” card slides into the lane beside them. A cyan loop arrow then traces once round the ring and back to its start, with “Same test again” beside it, and an “At least quarterly” tag settles under the loop. No result or winner appears.
An illustrative contract question-answering job meets its deciding constraint, a benchmark earns a test, a short-context checkpoint is recorded as a fit issue, candidates run under fixed conditions and the test is set to re-run. No model is chosen.
Read the article: Best Arabic LLM in 2026: Which Model for Which Job (Falcon, ALLaM, Jais, Fanar)