Hollow markers & dashed spokes: axis not yet scored
G
85.5
A
98
U
—
N
—
T
n/d
L
—
E
—
T
48.3
Specification
- Parameters
- 35B total (MoE; active-per-token not confirmed in model card)
- Architecture
- qwen3_5_moe
- Size on disk
- 20.4 GB
- Quantization
- 4bit
- Format
- MLX
- Class
- Standard
- Reasoning (CoT)
- Yes — emits reasoning tokens
- Internal ID
- M28
- Mean speed
- 115.6 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 222.3 / 260 17.1 13/13 118.9 run 2026-08-22
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| A1 | Dev/Ops Scripting | 14/20 | 117.5 | |
| A2 | Dev/Ops Scripting | 14/20 | 124 | |
| A3 | Dev/Ops Scripting | 19/20 | 121.6 | |
| A4 | Dev/Ops Scripting | 19/20 | 121.4 | |
| A5 | Dev/Ops Scripting | 19/20 | 128.3 | |
| A6 | Dev/Ops Scripting | 20/20 | 121.1 | |
| B1 | Document Processing | 19/20 | 123.8 | |
| B2 | Document Processing | 20/20 | 121.5 | |
| B3 | Document Processing | 0/20 | — | |
| C1 | Content Production | 20/20 | 112.2 | |
| C2 | Content Production | 19/20 | 109.7 | |
| C3 | Content Production | 19/20 | 111.5 | |
| C4 | Content Production | 20/20 | 113.7 |
Category mean Dev/Ops Scripting 17.5Document Processing 13Content Production 19.5
Agentic tool-calling & protocol adherence 1b 156.8 / 160 19.6 8/8 112.3 run 2026-08-22
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| WA1 | Tool Calling & Protocol Adherence | 20/20 | 109.3 | |
| WA2 | Tool Calling & Protocol Adherence | 20/20 | 99.3 | |
| WA3 | Tool Calling & Protocol Adherence | 20/20 | 114.1 | |
| WA4 | Tool Calling & Protocol Adherence | 20/20 | 83 | |
| WA5 | Tool Calling & Protocol Adherence | 20/20 | 115.3 | |
| WB1 | Agent Robustness & State Management | 20/20 | 127.3 | |
| WB2 | Agent Robustness & State Management | 20/20 | 126.8 | |
| WB3 | Agent Robustness & State Management | 17/20 | 123.4 |
Category mean Tool Calling & Protocol Adherence 20Agent Robustness & State Management 19
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.