Model benchmarks
swift-1.5-iq3_xxs local LLM performance
As of November 2026, swift-1.5-iq3_xxs runs at up to 116.5 tok/s for local inference (best of 13 community benchmark runs across 6 GPUs).
Model size
Unknown
Peak speed
116.5 tok/s
Average speed
62.8 tok/s
Avg PP
316.6 tok/s
Min memory
3.4 GB
Max context
262,144 tokens
Avg output / run
33,393 tokens
Avg runtime / run
9m 36s
Avg quality
84.1
Benchmark runs
13
GPUs tested
6
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 13 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 | 78.8 | 84.1 | 88.3 |
| Agent Workflow | 72.7 | 84.6 | 90.9 |
| Code Generation | 61.0 | 75.9 | 83.2 |
| Role Play & Narrative | 83.9 | 90.0 | 95.1 |
| Research & Analysis | 81.5 | 85.8 | 89.4 |
Performance by hardware and tool
Every hardware/tool/quantization combination swift-1.5-iq3_xxs has been benchmarked on, ranked by peak token generation speed. Last updated November 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | Strata | — | 116.5 tok/s | 116.5 tok/s | 3.4 GB | 262,144 tokens | 85.9 | 1 |
| NVIDIA GeForce RTX 3090 | Strata | — | 107.8 tok/s | 105.9 tok/s | 65.2 GB | 65,536 tokens | 79.4 | 2 |
| AMD Radeon AI PRO R9700 | Strata | — | 90.1 tok/s | 90.1 tok/s | n/a | 65,536 tokens | 88.5 | 1 |
| AMD Radeon RX 9070/9070 XT/9070 GRE | Strata v0.2.6 | — | 57.2 tok/s | 57.2 tok/s | 57.0 GB | 65,536 tokens | 88.2 | 1 |
| AMD Radeon RX 7900 XTX | Strata | — | 55.5 tok/s | 53.1 tok/s | n/a | 65,536 tokens | 84.0 | 5 |
| NVIDIA GeForce RTX 3070 Ti | polystrata | — | 26.5 tok/s | 25.1 tok/s | 10.1 GB | 65,536 tokens | 84.0 | 3 |
Benchmark runs
All 13 swift-1.5-iq3_xxs runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is swift-1.5-iq3_xxs good for coding?
- In our benchmarks, swift-1.5-iq3_xxs scores 75.9/100 for coding. It runs at about 62.8 tok/s, so if you want more speed, qwen3.8-27b-Swift1.5-Uncensored-ATX-AI3 is faster (~85.5 tok/s) and still scores well for coding (84.8/100).
- Is swift-1.5-iq3_xxs good for agentic (tool-using) tasks?
- In our benchmarks, swift-1.5-iq3_xxs scores 84.6/100 for agentic workflows. It runs at about 62.8 tok/s, so if you want more speed, Nex-N2.5-mini-APEX-Mini is faster (~88.2 tok/s) and still scores well for agentic workflows (91.3/100).
- How fast is swift-1.5-iq3_xxs for local inference?
- Across 13 community benchmark runs, swift-1.5-iq3_xxs reaches up to 116.5 tok/s and averages 62.8 tok/s, with the fastest results on NVIDIA GeForce RTX 5090.
- How much memory does swift-1.5-iq3_xxs need?
- The leanest observed configuration used about 3.4 GB of memory.
- Which tools have been used to run swift-1.5-iq3_xxs?
- Benchmarks were submitted using Strata, Strata v0.2.6, polystrata. Results are community-contributed and updated as new runs arrive.