Model benchmarks
K2-Horizon-MoVA-36B-A4B-MLX-4bit local LLM performance
As of September 2026, K2-Horizon-MoVA-36B-A4B-MLX-4bit runs at up to 49.1 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
36B
Peak speed
49.1 tok/s
Average speed
48.8 tok/s
Avg prefill
454.7 tok/s
Min memory
23.2 GB
Max context
524,288 tokens
Avg output / run
43,587 tokens
Avg runtime / run
32m 49s
Avg quality
84.3
Benchmark runs
2
GPUs tested
1
Quality by task
Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 2 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 83.9 | 84.3 | 84.6 |
| Agent Workflow | 90.6 | 90.9 | 91.1 |
| Code Generation | 80.5 | 85.3 | 90.1 |
| Role Play & Narrative | 76.8 | 79.6 | 82.4 |
| Research & Analysis | 78.2 | 81.4 | 84.5 |
Performance by hardware and tool
Every hardware/tool/quantization combination K2-Horizon-MoVA-36B-A4B-MLX-4bit has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Apple M5 Max | oMLX | — | 49.1 tok/s | 48.8 tok/s | 23.2 GB | 524,288 tokens | 84.3 | 2 |
Frequently asked questions
- Is K2-Horizon-MoVA-36B-A4B-MLX-4bit good for coding?
- In our benchmarks, K2-Horizon-MoVA-36B-A4B-MLX-4bit scores 85.3/100 for coding, currently the best-scoring model for coding that runs on consumer hardware (≤24 GB VRAM). It runs at about 48.8 tok/s, so if you want more speed, Qwen3.6-35B-A3B-Q4_K_XL is faster (~156.9 tok/s) and still scores well for coding (82.2/100).
- Is K2-Horizon-MoVA-36B-A4B-MLX-4bit good for agentic (tool-using) tasks?
- In our benchmarks, K2-Horizon-MoVA-36B-A4B-MLX-4bit scores 90.9/100 for agentic workflows, among the top 3 for agentic workflows on consumer hardware (≤24 GB VRAM). It runs at about 48.8 tok/s, so if you want more speed, unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:UD-Q4_K_XL is faster (~169.2 tok/s) and still scores well for agentic workflows (83.0/100).
- How fast is K2-Horizon-MoVA-36B-A4B-MLX-4bit for local inference?
- Across 2 community benchmark runs, K2-Horizon-MoVA-36B-A4B-MLX-4bit reaches up to 49.1 tok/s and averages 48.8 tok/s, with the fastest results on Apple M5 Max.
- How much memory does K2-Horizon-MoVA-36B-A4B-MLX-4bit need?
- The leanest observed configuration used about 23.2 GB of memory.
- Which tools have been used to run K2-Horizon-MoVA-36B-A4B-MLX-4bit?
- Benchmarks were submitted using oMLX. Results are community-contributed and updated as new runs arrive.