Comfortable on the FP32 / quantized profile.
- Model size
- 568M · 8K max
- Model format
- FP32 / quantized
- Runs with
- FlagEmbedding
The official example runs on CPU; throughput still depends on pair count and sequence length.
Use AI.Or get paid to run it.
Simulated marketOpen playground
Simulated market
8 requesters · 10 providers
8 requesters · 10 providers. Gateways create demand. Providers compete to serve it.
Illustrative market activity, not live network data. Eight requester gateways and ten compute providers participate in three example transactions. Providers compete on price, speed, model availability, and evidence. Selected providers stream tokens back and both sides co-sign the resulting receipt.
Choose AI models from the network and pay only for what you use. Or connect your own computer and earn when it runs models for someone else.
No high-end GPU needed. CPUs and older graphics cards can earn with smaller AI models.
Built for the open model ecosystem
Open model ecosystem organizations represented here: OpenAI, Meta AI, Qwen, Mistral AI, DeepSeek, Google DeepMind, Z.ai, Moonshot AI, MiniMax, NVIDIA, Black Forest Labs, Stability AI, Tongyi-MAI, Tencent Hunyuan, Lightricks, Microsoft, Resemble AI, Hexgrad, ACE-Step, BAAI, Jina AI, Nomic AI, Hugging Face. Their logos identify their respective organizations and do not imply partnership or endorsement.
Why OpenMayhem
You set the rules.
Choose what a request may cost, how fast it should run, and what protection it needs. OpenMayhem finds a matching provider and records what happened.
Set your limits
Choose your maximum price, minimum speed, and level of protection. If no provider matches, the request stops before you are charged.
Your price stays locked
Prices update hourly with supply and demand. Once your request starts, the agreed price cannot change.
Proof for every request
Every completed request creates a signed record of its usage and cost, so you have evidence if something goes wrong.
Choose who you trust
Use community providers, require proof that a machine is real, keep prompts hidden from the operator, or use only verified businesses.
Compare protection optionsHardware checker
What can your computer run?
Choose your hardware. We’ll show you which AI models are a good fit.
Your matches
20 models should run
17 comfortably · 3 with lighter settings · 2 need a local test.
Want to be certain? mayhem doctor checks your exact setup before anything downloads.
Privacy & payment
Choose what matters to you.
Choose signed records, verified hardware, confidential processing, or a verified business for each request.
Pick the protection you need
Choose for each request. More is not always better.
- Usage recordsT1Signed records of what ran and what it cost.Optional
- Verified hardwareT2Confirms the request runs on a verified machine.Optional
- Confidential processingT3Keeps prompts private while they are processed. selected
- Verified providerT4Confirms the business operating the service.Optional
Your prompt, your choice
Always encrypted
Protected while it travels.
Confidential with T3
Processed inside protected hardware.
Simple, visible fees
Card example · $1.0085¢
to the provider
15¢
network fee
Don't take our word for it. The rules and payment code are public.
Common questions
What you need to know.
Straight answers about paying, privacy, compatibility, and earning with your computer.
Get started
Set up with a coding agent.
There is no one-click installer yet. Choose what you want, copy the prompt, and paste it into Codex, Claude Code, or another coding agent.
Earn with your machine
The agent prepares your machine to run AI models for other people.
- 01
Check your hardware
Reads your GPU and available memory.
- 02
Match AI models
Finds models your hardware can run.
- 03
Start providing
Launches provider mode and checks it.
Prefer to install it yourself? Read the setup guide (opens in a new tab) · Questions? Ask the community (opens in a new tab) · Join Telegram (opens in a new tab)