Hollow markers & dashed spokes: axis not yet scored
G
67.5
A
92
U
—
N
—
T
n/d
L
—
E
—
T
23.5
Specification
- Parameters
- 12.19B
- Architecture
- gemma3
- Size on disk
- 7.51 GB
- Quantization
- 4bit (QAT)
- Format
- MLX
- Class
- Standard
- Reasoning (CoT)
- No
- Internal ID
- M32
- Mean speed
- 55.9 tok/s across suites
- Model card
- huggingface.co
Suite results
Suite Score Avg / 20 Tests tok/s Run
General capability (13-task real-workload suite) 1a 175.5 / 260 13.5 13/13 58.1 run 2026-09-06
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| A1 | Dev/Ops Scripting | 8/20 | 63 | |
| A2 | Dev/Ops Scripting | 12/20 | 60.9 | |
| A3 | Dev/Ops Scripting | 9/20 | 56.9 | |
| A4 | Dev/Ops Scripting | 9/20 | 55.9 | |
| A5 | Dev/Ops Scripting | 9/20 | 61.7 | |
| A6 | Dev/Ops Scripting | 20/20 | 62.8 | |
| B1 | Document Processing | 20/20 | 60.8 | |
| B2 | Document Processing | 20/20 | 61.5 | |
| B3 | Document Processing | 9/20 | 57.9 | |
| C1 | Content Production | 18/20 | 59.8 | |
| C2 | Content Production | 16/20 | 52.1 | |
| C3 | Content Production | 16/20 | 50.3 | |
| C4 | Content Production | 9/20 | 51.3 |
Category mean Dev/Ops Scripting 11.2Document Processing 16.3Content Production 14.8
Agentic tool-calling & protocol adherence 1b 147.2 / 160 18.4 8/8 53.7 run 2026-09-06
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| WA1 | Tool Calling & Protocol Adherence | 20/20 | 44.2 | |
| WA2 | Tool Calling & Protocol Adherence | 20/20 | 48.8 | |
| WA3 | Tool Calling & Protocol Adherence | 20/20 | 50.8 | |
| WA4 | Tool Calling & Protocol Adherence | 20/20 | 45 | |
| WA5 | Tool Calling & Protocol Adherence | 20/20 | 56.5 | |
| WB1 | Agent Robustness & State Management | 17/20 | 62.8 | |
| WB2 | Agent Robustness & State Management | 19/20 | 64.1 | |
| WB3 | Agent Robustness & State Management | 11/20 | 57.2 |
Category mean Tool Calling & Protocol Adherence 20Agent Robustness & State Management 15.7
Each test is LLM-judged 0–20 against a fixed rubric; suite max = tests × 20. Rows with a ▶ expand to the per-test breakdown. Where fewer tests ran than the suite total, unrun tests count as zero toward the axis score. See methodology.