Model benchmarks
qwen3.8-flash-next-iq3_s local LLM performance
As of November 2026, qwen3.8-flash-next-iq3_s runs at up to 190.3 tok/s for local inference (best of 19 community benchmark runs across 10 GPUs).
Model size
Unknown
Peak speed
190.3 tok/s
Average speed
89.8 tok/s
Avg PP
446.2 tok/s
Min memory
3.4 GB
Max context
262,144 tokens
Avg output / run
44,718 tokens
Avg runtime / run
10m 21s
Avg quality
81.6
Benchmark runs
19
GPUs tested
10
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 19 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 | 68.9 | 81.6 | 88.6 |
| Agent Workflow | 72.1 | 84.6 | 94.2 |
| Code Generation | 19.9 | 64.7 | 81.9 |
| Role Play & Narrative | 86.5 | 91.3 | 96.2 |
| Research & Analysis | 79.4 | 85.6 | 91.2 |
Performance by hardware and tool
Every hardware/tool/quantization combination qwen3.8-flash-next-iq3_s 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 | — | 190.3 tok/s | 141.2 tok/s | 3.4 GB | 262,144 tokens | 85.2 | 4 |
| NVIDIA GeForce RTX 3080 | Strata | — | 120.3 tok/s | 117.4 tok/s | 114.7 GB | 262,144 tokens | 82.8 | 2 |
| NVIDIA GeForce RTX 3090 Ti | Strata | — | 115.6 tok/s | 115.6 tok/s | 68.9 GB | 131,072 tokens | 83.7 | 1 |
| NVIDIA GeForce RTX 3090 | Strata | — | 89.2 tok/s | 79.1 tok/s | 62.0 GB | 163,840 tokens | 79.5 | 4 |
| AMD Radeon AI PRO R9700 | Strata | — | 77.7 tok/s | 77.7 tok/s | n/a | 131,072 tokens | 86.6 | 1 |
| AMD Radeon RX 9070/9070 XT/9070 GRE | Strata | — | 67.9 tok/s | 66.4 tok/s | 62.0 GB | 65,536 tokens | 87.1 | 2 |
| AMD Radeon RX 7900 XTX | Strata | — | 67.3 tok/s | 67.3 tok/s | 98.1 GB | 32,768 tokens | 69.9 | 1 |
| AMD Radeon RX 7900 XT | Strata | — | 63.3 tok/s | 60.3 tok/s | 67.5 GB | 65,536 tokens | 82.3 | 2 |
| Quadro RTX 5000 with Max-Q Design | strata 0.2.9 / engine 0.1.55 | — | 46.5 tok/s | 46.5 tok/s | 67.5 GB | 8,192 tokens | 70.3 | 1 |
| NVIDIA GeForce RTX 4070 | llama.cpp | — | 29.7 tok/s | 29.7 tok/s | 51.5 GB | 32,768 tokens | 76.0 | 1 |
Benchmark runs
All 19 qwen3.8-flash-next-iq3_s runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is qwen3.8-flash-next-iq3_s good for coding?
- In our benchmarks, qwen3.8-flash-next-iq3_s scores 64.7/100 for coding. It runs at about 89.8 tok/s, so if you want more speed, ornith-1.5-35b-a3b-apex-mtp-i-quality is faster (~105.2 tok/s) and still scores well for coding (83.6/100).
- Is qwen3.8-flash-next-iq3_s good for agentic (tool-using) tasks?
- In our benchmarks, qwen3.8-flash-next-iq3_s scores 84.6/100 for agentic workflows. It runs at about 89.8 tok/s, so if you want more speed, cyber-tiel-coder-35b-a3b-mtp is faster (~115.9 tok/s) and still scores well for agentic workflows (86.4/100).
- How fast is qwen3.8-flash-next-iq3_s for local inference?
- Across 19 community benchmark runs, qwen3.8-flash-next-iq3_s reaches up to 190.3 tok/s and averages 89.8 tok/s, with the fastest results on NVIDIA GeForce RTX 5090.
- How much memory does qwen3.8-flash-next-iq3_s need?
- The leanest observed configuration used about 3.4 GB of memory.
- Which tools have been used to run qwen3.8-flash-next-iq3_s?
- Benchmarks were submitted using Strata, llama.cpp, strata 0.2.9 / engine 0.1.55. Results are community-contributed and updated as new runs arrive.