Compare runs
Pick two or more runs to compare. Qwen3.8-27B on M4 Max 128 GB is the first; add another below.
| Run | Qwen3.8-27BM4 Max 128 GB · llama.cpp llama-swap (version not exposed)Take out Qwen3.8-27B |
|---|---|
| Score | 1,950Silver |
| Tasks passed | 41 of 44 |
| Points | 57 of 64 |
| Speed | 21.9 tok/s |
| Tokens | 270,902 |
| Summarizing | 5/5 |
| One-sentence summary | Passed23.9 tok/s |
| Meeting notes in three bullets | Passed18.8 tok/s |
| Review sentiment and complaint | Passed14.7 tok/s |
| Faithful summary of a study | Passed19.8 tok/s |
| Action items from an email thread | Passed23.1 tok/s |
| Extracting | 5/5 |
| Fields from an email as JSON | Passed24.5 tok/s |
| Every email address, once | Passed22.5 tok/s |
| A list as CSV | Passed14.9 tok/s |
| The real deadline | Passed16 tok/s |
| Warnings and errors from a log | Passed15.7 tok/s |
| Following instructions | 5/5 |
| Exactly one word | Passed21.8 tok/s |
| Three lowercase lines | Passed22.7 tok/s |
| Exact JSON shape | Passed24.7 tok/s |
| Sort a list | Passed23.8 tok/s |
| A name with three rules | Passed23.3 tok/s |
| Reasoning | 5/5 |
| Unit price | Passed23.6 tok/s |
| Days of the week | Passed24.2 tok/s |
| Who owns the cat | Passed23.1 tok/s |
| Arrival time | Passed25.2 tok/s |
| Filling a tank | Passed24.8 tok/s |
| Reading code | 4/4 |
| What does it print | Passed24.3 tok/s |
| Trace a loop | Passed25 tok/s |
| Find the bug | Passed23.4 tok/s |
| Run a query in your head | Passed25.1 tok/s |
| Using tools | 5/5 |
| Call one tool | Passed23.6 tok/s |
| Don't call a tool when none is needed | Passed23.4 tok/s |
| Two calls for two answers | Passed16.2 tok/s |
| Chain two tools | Passed15.8 tok/s |
| Handle a failed call honestly | Passed18.8 tok/s |
| Building pages | 4/5 |
| A small website | Passed22.4 tok/s |
| A pricing table with a toggle | Passed20.6 tok/s |
| An accessible sign-up form | Passed23.2 tok/s |
| A to-do app that remembers | Passed22.8 tok/s |
| A bar chart in SVG | Failed22.5 tok/s |
| Building games | 0/2 |
| Build 2048 | Failed21.7 tok/s |
| Build Snake | Failed22.4 tok/s |
| Writing code | 3/3 |
| Merge intervals | Passed23.1 tok/s |
| An LRU cache | Passed23 tok/s |
| Evaluate arithmetic | Passed21.7 tok/s |
| Fixing code | 2/2 |
| Fix the bookings module | Passed23.6 tok/s |
| Fix the CSV reader | Passed24.1 tok/s |
| Planning and data | 3/3 |
| A query over hidden data | Passed22.7 tok/s |
| Schedule five meetings | Passed22.8 tok/s |
| A replacement in ten calls or fewer | Passed23.7 tok/s |
Scrolls sideways to show every run.
A check marks the best of each row. tok/s is tokens a second: how fast the model writes. How the score works.