Strata artifacts and availability boundaries were verified from primary sources.
Strata runs Qwen3.8-Flash-Next on 12 GB gaming GPUs
2026-09-25Frontier watchTRIAL
Strata is an MIT-licensed inference engine that distributes Qwen3.8-Flash-Next experts across GPU, RAM, CPU and SSD. It exposes OpenAI- and Anthropic-compatible local endpoints and documents a 12 GB GPU, 32 GB RAM and about 80 GB SSD floor. Author benchmarks report 60–94 output tokens per second on recent gaming GPUs.
What this means for you
For teams evaluating very large sparse models locally, Strata offers a concrete heterogeneous-memory serving design and agent-compatible APIs. Rapid repository adoption warrants a trial, but self-reported speed and aggressive quantization require quality and restart tests.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| License | MIT |
|---|---|
| Hardware Floor | 12 GB supported GPU, 32 GB RAM, ~80 GB SSD |
| Interfaces | OpenAI- and Anthropic-compatible APIs |
| Author Speed | 94 tok/s RTX 5070 Q2_0; 60 tok/s RX 9070 XT |
| Adoption | about 9.1k GitHub stars and 831 forks at review |
Technical impact 3
Engineering value 4
Adoption signal 5
Maturity 2
Verification cost 3
Risk 3