Model benchmarks
flash-next local LLM performance
As of November 2026, flash-next runs at up to 88.8 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
Unknown
Peak speed
88.8 tok/s
Average speed
82.1 tok/s
Avg PP
621.4 tok/s
Min memory
n/a
Max context
65,536 tokens
Avg output / run
80,859 tokens
Avg runtime / run
17m 42s
Avg quality
87.7
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 | 86.8 | 87.7 | 88.6 |
| Agent Workflow | 82.9 | 86.8 | 90.6 |
| Code Generation | 78.6 | 79.1 | 79.5 |
| Role Play & Narrative | 95.2 | 95.3 | 95.3 |
| Research & Analysis | 88.8 | 89.8 | 90.7 |
Performance by hardware and tool
Every hardware/tool/quantization combination flash-next 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon AI PRO R9700 | Strata | — | 88.8 tok/s | 82.1 tok/s | n/a | 65,536 tokens | 87.7 | 2 |
Benchmark runs
All 2 flash-next runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is flash-next good for coding?
- In our benchmarks, flash-next scores 79.1/100 for coding. It runs at about 82.1 tok/s, so if you want more speed, swift-1.5-27b is faster (~96.8 tok/s) and still scores well for coding (84.8/100).
- Is flash-next good for agentic (tool-using) tasks?
- In our benchmarks, flash-next scores 86.8/100 for agentic workflows. It runs at about 82.1 tok/s, so if you want more speed, Ornith-1.5-35B-A3B-BigBang-oQ8e-mtp is faster (~102.7 tok/s) and still scores well for agentic workflows (91.1/100).
- How fast is flash-next for local inference?
- Across 2 community benchmark runs, flash-next reaches up to 88.8 tok/s and averages 82.1 tok/s, with the fastest results on AMD Radeon AI PRO R9700.
- Which tools have been used to run flash-next?
- Benchmarks were submitted using Strata. Results are community-contributed and updated as new runs arrive.