Models tested against my personal workflows, rather than an arbitrary benchmark.
Latest result per task for each model. Full stack averages FS1 and FS2, and Lead averages L1 and L2; each needs both tasks; with one missing it shows as incomplete. Click a row for task scores, dates and source runs; click a column header to sort.
The effective token rate sets how much you value time saved, and the Best value pick in each table is recommended on that basis. At a lower rate, price counts for more, so cheaper models win even when they are slower. At a higher rate, speed counts for more, so faster models win even when they cost more.
Eight questions across six repositories, answered with exact paths and line numbers.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Compress a 3,853-line business-rule spec and a 723-line client audit, in under 40 lines each.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Name every file a known change has to touch. The real fixes touched 25 and 19 files.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Ship a CMS section end to end, and apply a design batch that then gets corrected mid-task.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Build a slice of a shipped warehouse feature (FS1), and repair a workout log's history paging and monthly totals (FS2).
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Build a terminal download dashboard, with a fixed recovery follow-up only if the first saved implementation does not already satisfy the complete checklist.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Make a slow, leaky Astro + Sanity Dockerfile fast and safe, and fix a broken server from a copy of its files.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Three storefront-header and cart/menu diffs: the header change with two injected requirement violations, the shipped header with two unmet requirements, and a mobile fix with a menu-view mismatch.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|
Best value: the paid model within 8 points of the best paid score, scoring 75 or above, with the lowest cost once its time is charged at the rate set above.
Run a scripted client conversation as the orchestrator: a vague brief that changes midway and a question that needs a lookup (L1), then a pure recommendation followed by a lookup (L2). Scout, digest and build are pinned, so only the lead changes.
| Model | Score | Cost | Effective | Time | Tokens | Run |
|---|