VPS Arena

Run gpt-oss-20b (MoE) on a VPS.

September 2026 · 138 of 280 plans fit · re-ranked daily

OpenAI's open-weight reasoner ships in MXFP4: ~13 GB on disk, 3.6B active, adjustable reasoning effort. Fits a 16 GB VPS. On a CPU-only VPS it needs about 14 GB of RAM: 12.1 GB of weights in the model's native format, 410 MB of KV cache for an 8,192-token context and 1.5 GB for the OS and runtime. 138 of the 280 plans in our index fit; the cheapest comfortable pick is Contabo's Cloud VPS 8 · 8 vCPU · 24 GB at $16.27/mo, streaming an estimated 4.4–8.7 tok/s.

RAM needed · CPU inference

14 GB

weights · native format
12.1 GB
KV cache · 8K context
410 MB
OS + runtime headroom
1.5 GB

Model card

Size
21B total · 3.6B active
Context
128K tokens
Kind
reasoning
Released
2025-08
License
Apache-2.0

Facts fetched from Hugging Face on 3 Sept 2026: exact GGUF file sizes, KV geometry from the GGUF header · 6,519,052 downloads.

ollama run gpt-oss:20bmodel card ↗OpenAI

Best VPS plans for gpt-oss-20b (MoE)

ranked by estimated tokens/s per dollar, comfortable fits first

  1. 1
    ContaboCloud VPS 8 · 8 vCPU · 24 GB

    8 vCPU shared · 24 GB RAM · 300 GB SSD

    runs comfortably~4.4–8.7 tok/s$16.27/mo
  2. 2
    netcupRS 4000 G12 · 12 dedicated cores · 32 GB

    12 vCPU dedicated · 32 GB RAM · 1024 GB NVME

    runs comfortably~8.2–16 tok/s$38.99/mo
  3. 3
    ContaboCloud VPS 12 · 12 vCPU · 48 GB

    12 vCPU shared · 48 GB RAM · 400 GB SSD

    runs comfortably~5.4–11 tok/s$29.05/mo
  4. 4
    netcupVPS 4000 G12 · 12 vCore · 32 GB

    12 vCPU shared · 32 GB RAM · 1024 GB NVME

    runs comfortably~5.4–11 tok/s$31.66/mo
  5. 5
    HetznerCX53 · 16 vCPU · 32 GB

    16 vCPU shared · 32 GB RAM · 320 GB NVME

    runs comfortably~5.4–11 tok/s$34.27/mo
  6. 6
    OVHcloudVPS-4 2027 · 8 vCPU / 24 GB

    8 vCPU shared · 24 GB RAM · 200 GB NVME

    runs comfortably~4.4–8.7 tok/s$27.50/mo

Speed = effective memory bandwidth ÷ active weight bytes (4 GB/s per shared vCPU, 6 per dedicated, ×0.5 for MoE routing overhead), shown as a band. Real numbers depend on the host CPU generation, AVX-512/AMX support and how noisy the neighbours are — treat these as order-of-magnitude.

Or buy hardware · 13 reference machines fit

all machines →
  • Raspberry Pi 5 (16 GB)16 GB · 17 GB/s · ~1–1.6 tok/s$305= 19 mo of VPS
  • GeForce RTX 5060 Ti 16 GB (card only)16 GB · 448 GB/s · ~53–82 tok/s$805*= 50 mo of VPS
  • GeForce RTX 5070 Ti 16 GB (card only)16 GB · 896 GB/s · ~100+ tok/s$1,100*= 68 mo of VPS
  • Mac mini (M6, 32 GB)32 GB · 153 GB/s · ~18–28 tok/s$1,299= 80 mo of VPS
  • GeForce RTX 5080 16 GB (card only)16 GB · 960 GB/s · ~100+ tok/s$1,500*= 92 mo of VPS
  • Mac mini (M5 Pro, 24 GB)24 GB · 307 GB/s · ~37–56 tok/s$1,699= 104 mo of VPS

Buy · per month

$8.78

$8.47 hardware + $0.31 power

Rent · per month

$16.27

Contabo Cloud VPS 8 · 8 vCPU · 24 GB

Break-even

The Raspberry Pi 5 (16 GB) pays for itself after 19 months of replacing the VPS — and it streams an estimated 1–1.6 tok/s against the VPS's CPU-only pace.

* approximate: August 2026 US retail median rather than list price. Local speed = peak bandwidth × 0.7 efficiency (0.35 on CPU-only boards) ÷ active weight bytes; unified-memory machines are assumed to give models 75% of their RAM. GPU cards need a host PC that is not included in the price.

Frequently asked

How much RAM does gpt-oss-20b (MoE) need?
About 14 GB for CPU inference at Q4_K_M: 12.1 GB of weights, 410 MB of KV cache at 8,192 tokens of context, and 1.5 GB of headroom. Longer contexts need more KV cache (48 KB per token for this model).
What is the cheapest VPS that can run gpt-oss-20b (MoE)?
Contabo Cloud VPS 8 · 8 vCPU · 24 GB (24 GB RAM, 8 vCPU) at $16.27/mo excl. VAT runs it comfortably as of 5 Sept 2026.
How fast will gpt-oss-20b (MoE) run on a VPS without a GPU?
Roughly 4.4–8.7 tok/s on the top pick. Token generation is bound by memory bandwidth — each token streams the active-expert weights once — so more and dedicated vCPUs help; a GPU or Apple-silicon machine is 10–50× faster.
Same lineupgpt-oss-120b (MoE)Same size classLlama 4 Scout (109B MoE)Qwen3 30B-A3B (MoE)VPS for AI agents (API-hosted models) →