Model benchmarks
Ornith-1.5-35B-A3B-MLX-4bit local LLM performance
As of October 2026, Ornith-1.5-35B-A3B-MLX-4bit runs at up to 84.6 tok/s for local inference (best of 4 community benchmark runs across 1 GPU).
Model size
35B
Peak speed
84.6 tok/s
Average speed
83.2 tok/s
Avg PP
2093.3 tok/s
Min memory
18.8 GB
Max context
262,144 tokens
Avg output / run
39,090 tokens
Avg runtime / run
8m 32s
Avg quality
74.5
Benchmark runs
4
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 4 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 | 69.5 | 74.5 | 82.3 |
| Agent Workflow | 27.3 | 65.7 | 82.9 |
| Code Generation | 31.7 | 64.5 | 84.8 |
| Role Play & Narrative | 72.1 | 81.3 | 86.1 |
| Research & Analysis | 83.3 | 86.5 | 88.9 |
Performance by hardware and tool
Every hardware/tool/quantization combination Ornith-1.5-35B-A3B-MLX-4bit has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Apple M4 Pro | oMLX | — | 84.6 tok/s | 84.6 tok/s | 18.8 GB | 65,536 tokens | 75.0 | 1 |
| Apple M4 Pro | oMLX | 4bit | 84.1 tok/s | 82.7 tok/s | 19.3 GB | 262,144 tokens | 74.3 | 3 |
Benchmark runs
All 4 Ornith-1.5-35B-A3B-MLX-4bit runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Ornith-1.5-35B-A3B-MLX-4bit good for coding?
- In our benchmarks, Ornith-1.5-35B-A3B-MLX-4bit scores 64.5/100 for coding. It runs at about 83.2 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e is faster (~97.4 tok/s) and still scores well for coding (80.9/100).
- Is Ornith-1.5-35B-A3B-MLX-4bit good for agentic (tool-using) tasks?
- In our benchmarks, Ornith-1.5-35B-A3B-MLX-4bit scores 65.7/100 for agentic workflows. It runs at about 83.2 tok/s, so if you want more speed, qwen3.8-flash-next-iq3_s is faster (~100.4 tok/s) and still scores well for agentic workflows (90.1/100).
- How fast is Ornith-1.5-35B-A3B-MLX-4bit for local inference?
- Across 4 community benchmark runs, Ornith-1.5-35B-A3B-MLX-4bit reaches up to 84.6 tok/s and averages 83.2 tok/s, with the fastest results on Apple M4 Pro.
- How much memory does Ornith-1.5-35B-A3B-MLX-4bit need?
- The leanest observed configuration used about 18.8 GB of memory (quantizations tested: 4bit).
- Which tools have been used to run Ornith-1.5-35B-A3B-MLX-4bit?
- Benchmarks were submitted using oMLX. Results are community-contributed and updated as new runs arrive.