2026-10-11 · Mac · Apple M5 Max
A model scorecard round on M5 Max · Apple M5 Max (integrated GPU) · 128 GB: 4 models across 4 task sets. Every round is listed in the model scorecard, with its results side by side in one table.
M5 Max · Apple M5 Max (integrated GPU) · 128 GB · anthropic + anthropic-cli + codex-cli + mlx + openai · 1 trial per task · 2026-10-11 · gezel 985c646b2e13 · catalog 0.1.94
Compares Claude Sonnet 5.5 and GPT-6 Luna, each reached two ways (through its maker's API and through its command-line app), with two models running on the Mac itself. Generalist mode was on. The run stopped early because of a fault in our test equipment, so a few tasks never ran. Some failures came from our tests or from how gezel connected to the command-line apps, not from the models, and are not counted against them. A model without a result for every task in a set is left out of that table rather than counted as passing. The command-line apps were tested again on October 12 after those problems were fixed.
General capability
| Model | Size | Tasks passed | Performance |
|---|---|---|---|
| Claude Sonnet 5.5 · Anthropic API | cloud | 11/11 (some tasks run once — count not rate) | — |
| GPT-6 Luna · OpenAI API | cloud | 11/11 (some tasks run once — count not rate) | — |
| Qwen 3.8 27B Q4 | 27B | 10/11 (some tasks run once — count not rate) | — |
Not published — some tasks could not be measured on this round: GPT-6 Luna · Codex CLI (1 task(s) unmeasured); Claude Sonnet 5.5 · Claude CLI (2 task(s) unmeasured); Gemma 4 31B Q4 (1 task(s) unmeasured)
Tasks in this set: 11.
Office and knowledge work
| Model | Size | Tasks passed | Performance |
|---|---|---|---|
| Claude Sonnet 5.5 · Anthropic API | cloud | 13/13 (some tasks run once — count not rate) | — |
| GPT-6 Luna · OpenAI API | cloud | 13/13 (some tasks run once — count not rate) | — |
| Gemma 4 31B Q4 | 31B | 11/13 (some tasks run once — count not rate) | — |
| Qwen 3.8 27B Q4 | 27B | 11/13 (some tasks run once — count not rate) | — |
Not published — some tasks could not be measured on this round: GPT-6 Luna · Codex CLI (4 task(s) unmeasured); Claude Sonnet 5.5 · Claude CLI (5 task(s) unmeasured)
Tasks in this set: 13.
Engineering work
| Model | Size | Tasks passed | Performance |
|---|---|---|---|
| Claude Sonnet 5.5 · Anthropic API | cloud | 10/10 (some tasks run once — count not rate) | — |
| GPT-6 Luna · Codex CLI | cloud | 10/10 (some tasks run once — count not rate) | — |
| GPT-6 Luna · OpenAI API | cloud | 10/10 (some tasks run once — count not rate) | — |
| Qwen 3.8 27B Q4 | 27B | 9/10 (some tasks run once — count not rate) | — |
| Gemma 4 31B Q4 | 31B | 7/10 (some tasks run once — count not rate) | — |
Not published — some tasks could not be measured on this round: Claude Sonnet 5.5 · Claude CLI (1 task(s) unmeasured)
Tasks in this set: 10.
Complex workflows
Not published — some tasks could not be measured on this round: Claude Sonnet 5.5 · Anthropic API (2 task(s) unmeasured); Gemma 4 31B Q4 (2 task(s) unmeasured); GPT-6 Luna · OpenAI API (2 task(s) unmeasured); Qwen 3.8 27B Q4 (2 task(s) unmeasured); GPT-6 Luna · Codex CLI (3 task(s) unmeasured); Claude Sonnet 5.5 · Claude CLI (4 task(s) unmeasured)
Tasks in this set: 9.