Hollow markers & dashed spokes: axis not yet scored
Specification
- Parameters
- 30B total / ~3B active per token
- Architecture
- nemotron3_nano (hybrid Mamba2/attention MoE)
- Size on disk
- 17.79 GB
- Quantization
- 4bit
- Format
- MLX
- Runtime
- LM Studio
- Class
- Standard
- Reasoning (CoT)
- Yes — emits reasoning tokens
- Internal ID
- M14
- Mean speed
- 137.2 tok/s across suites
- Stall census
- 1 stall in 30 observed tests (3.3%)
- Reasoning appetite
- 1,413 tokens mean · 16,383 max
- Model card
- huggingface.co
Suite results
General capability (13-task real-workload suite) 1a 204.1 / 260 15.7 13/13 n=1 142.6 run 2026-07-27
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| A1 | Dev/Ops Scripting | 11/20 | 103.4 | |
| A2 | Dev/Ops Scripting | 18/20 | 137.2 | |
| A3 | Dev/Ops Scripting | 11/20 | 124.4 | |
| A4 | Dev/Ops Scripting | 16/20 | 113.7 | |
| A5 | Dev/Ops Scripting | 18/20 | 142 | |
| A6 | Dev/Ops Scripting | 0/20 | 112.7 | |
| B1 | Document Processing | 20/20 | 109.6 | |
| B2 | Document Processing | 20/20 | 128.4 | |
| B3 | Document Processing | 17/20 | 141.1 | |
| C1 | Content Production | 18/20 | 150.9 | |
| C2 | Content Production | 18/20 | 151.3 | |
| C3 | Content Production | 19/20 | 153.6 | |
| C4 | Content Production | 18/20 | 152.6 |
Category mean Dev/Ops Scripting 12.3Document Processing 19Content Production 18.3
Agentic tool-calling & protocol adherence 1b 148 / 160 18.5 8/8 n=1 162.8 run 2026-07-28
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| WA1 | Tool Calling & Protocol Adherence | 20/20 | 17.3 | |
| WA2 | Tool Calling & Protocol Adherence | 20/20 | 83.4 | |
| WA3 | Tool Calling & Protocol Adherence | 20/20 | 118.4 | |
| WA4 | Tool Calling & Protocol Adherence | 20/20 | 77.2 | |
| WA5 | Tool Calling & Protocol Adherence | 20/20 | 111.6 | |
| WB1 | Agent Robustness & State Management | 17/20 | 156.1 | |
| WB2 | Agent Robustness & State Management | 20/20 | 155.8 | |
| WB3 | Agent Robustness & State Management | 11/20 | 142.1 |
Category mean Tool Calling & Protocol Adherence 20Agent Robustness & State Management 16
Long-context retrieval & synthesis 1g 120 / 120 20 6/6 n=1 140 run 2026-08-06
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| L1 | Long-Context — Needle Retrieval | 20/20 | 61.3 | |
| L2 | Long-Context — Grounded QA with Distractors | 20/20 | 62.7 | |
| L3 | Long-Context — Ops Log Reasoning | 20/20 | 64.4 | |
| L4 | Long-Context — Faithful Summarization | 20/20 | 81.6 | |
| L5 | Long-Context — Instruction Retention | 20/20 | 52.4 | |
| L6 | Long-Context — Full-Haystack QA | 20/20 | 41.7 |
Category mean Long-Context — Needle Retrieval 20Long-Context — Grounded QA with Distractors 20Long-Context — Ops Log Reasoning 20Long-Context — Faithful Summarization 20Long-Context — Instruction Retention 20Long-Context — Full-Haystack QA 20
Long-context multi-needle (MRCR) 1g2 15 / 60 5 3/3 n=1 103.3 run 2026-08-14
| Test | Category | Score | tok/s | |
|---|---|---|---|---|
| L7 | Long-Context — Multi-Needle Retrieval (MRCR) | 11/20 | 58.2 | |
| L8 | Long-Context — Multi-Needle Retrieval (MRCR) | 0/20 | 43.8 | |
| L9 | Long-Context — Multi-Needle Retrieval (MRCR) | 4/20 | 35.3 |
Each test is LLM-judged 0–20 against a fixed rubric; suite max = tests × 20. An ADJ chip marks a cell carrying a subjective adjustment from the maintainer; the judged score is the one shown, is unchanged, and remains what the axes, ranks and tie bands are computed from. Hover for the reason. 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.