Capy Flash is listed as Open Source. Check https://github.com/ValerioDolci/capy-flash for the current plan limits, as vendors change pricing frequently.

Capy Flash

Q: What is Capy Flash?

Capy Flash is a code AI tool listed on Use AI Tools. Capy Flash runs Qwen3.8-Flash-Next, an MoE model in NVFP4, and Qwen3.8-27B on two 16 GB consumer GPUs. It uses two-GPU tensor parallelism with experts offloaded to the CPU, so large MoE checkpoints fit on hardware a hobbyist can own.

Q: What is Capy Flash best for?

Capy Flash is commonly used for Hobbyists with two consumer GPUs, Engineers testing MoE serving without a datacentre card, Anyone offloading experts to CPU RAM. It supports English.

Code⚡ Advanced✅ Global access

Hobbyists with two consumer GPUsEngineers testing MoE serving without a datacentre cardAnyone offloading experts to CPU RAM

Open SourceEnglish

Capy Flash runs Qwen3.8-Flash-Next, an MoE model in NVFP4, and Qwen3.8-27B on two 16 GB consumer GPUs. It uses two-GPU tensor parallelism with experts offloaded to the CPU, so large MoE checkpoints fit on hardware a hobbyist can own.

Visit Website Compare

Community Rating

Be the first to rate Capy Flash. Your feedback helps other users decide.

🎯 Real-World Use Cases

Two-GPU MoE serving

Split a large MoE model across two 16 GB cards with tensor parallelism.

CPU expert offload

Keep experts in system RAM when they do not fit in VRAM.

Local throughput testing

Measure what consumer hardware can actually sustain for a given checkpoint.

⚖️ Honest Review

✅ Pros

+Targets realistic consumer hardware
+Combines tensor parallel with expert offload
+Open source

⚠️ Cons

−Tied to specific Qwen checkpoints
−C++ build and tuning required

📚

Guides and comparisons about code tools

Engineers working around a table with laptops and infrastructure diagrams

CodeOct 20, 2026