Active machines

Hardware becomes
capability.

SolarCompute connects different classes of compute hardware to one controlled network. Each machine is identified by its capabilities, matched with approved runtimes, and made available for workloads it is suited to run.

Run a workload → Become a Host →
How a machine joins the network
01
Connect

A Host machine securely connects to SolarCompute.

02
Discover

Hardware and available capabilities are identified.

03
Match

Capability is matched against approved network workloads.

04
Execute

Suitable work can be assigned through controlled execution paths.

Capability driven

The hardware does not define the workload. Its capabilities do.

SolarCompute uses hardware discovery and approved workload matching so different machine configurations can contribute where they are technically suitable.

01 Machine Connected hardware
→
02 Discover Hardware profile
→
03 Match Approved capability
→
04 Runtime Controlled environment
→
05 Workload Network execution
Built for more than one model

One network.
A growing model library.

SolarCompute is being built to support a broad range of AI and compute workloads. Model availability can grow as new runtimes are validated and compatible hardware joins the network.

01
AI

Language & reasoning

Language models for chat, reasoning, research, document processing and specialised AI workflows.

LLM / reasoning
02
IMG

Image & vision

Generative image, visual processing and supported GPU-accelerated creative workloads.

GPU / visual AI
03
ML

Training

Supported machine-learning training, fine-tuning and experimental model workflows matched to suitable hardware.

Training / ML
04
CPU

Compute

CPU processing, simulation, rendering and controlled containerised workloads beyond generative AI.

CPU / GPU compute
OpenAI on SolarCompute

Open models that
scale with the network.

SolarCompute can run open-weight AI models locally on compatible network machines. As more capable hardware joins the network, larger reasoning, safety and multimodal workloads can become available through the same distributed architecture.

Current network capability
OPENAI OPEN-WEIGHT

gpt-oss-20b

Local reasoning on capable network machines.

A compact open-weight reasoning model designed for efficient local inference. Its architecture makes advanced reasoning practical on machines with sufficient local memory, without requiring every request to leave the SolarCompute network.

MODEL CLASS Reasoning
PARAMETERS ~20B
MEMORY CLASS ~16 GB+
EXECUTION Local
As network hardware grows
01
More capable machines

Higher VRAM, system memory and compute capability.

02
Larger model classes

More demanding reasoning and specialist models become viable.

03
Broader workloads

The network expands without changing the way customers submit work.

Future hardware class
01 REASONING

gpt-oss-120b

A substantially larger open-weight reasoning model designed for high-memory accelerator hardware. As stronger SolarCompute machines come online, this represents the next class of local reasoning workload the network can support.

~120B parameters ~80 GB memory class
Safety architecture
02 GUARDRAILS

gpt-oss-safeguard

Safety-focused reasoning variants provide a path for policy-aware classification and custom safety controls, complementing SolarCompute's own workload validation and controlled execution guardrails.

20B + 120B variants Policy aware
Speech & audio
03 AUDIO

Whisper

OpenAI's speech-recognition model provides a natural path for transcription and audio-processing workloads running on compatible local or distributed compute.

Speech recognition Transcription
THE NETWORK EFFECT

Every stronger hardware class that joins SolarCompute can expand what the network is capable of running.

Model ecosystem

What's here today
is only the beginning.

SolarCompute separates physical hardware from the model and workload layer. That means the network can expand its supported software without every machine needing to be identical.

NOW
Available on the network

Approved models, runtimes and compute workloads matched to compatible machines.

NEXT
More models being integrated

Additional language, reasoning, image and specialist model families can move through validation and workload approval.

GROW
Broader capability

As network hardware expands, SolarCompute can support larger and more demanding model and compute classes.

Hardware matching

The right work goes to the right machine.

A connected machine is not treated as generic compute. SolarCompute uses its reported hardware profile and approved capability matching to determine which workload paths are suitable.

See how SolarCompute works →
01 Hardware CPU · GPU · memory · storage
↓
02 Capability What the machine can support
↓
03 Approved runtime Controlled software environment
↓
04 Model / workload Compatible network work
↓
05 Job Assigned for execution
Controlled network

Connected without being exposed.

SolarCompute can understand what a machine is capable of without publicly exposing sensitive Host, network or workload information.

Protected by design
01

Host identities and private machine names are not exposed.

02

Exact addresses and private network details remain protected.

03

Tokens, keys and internal identifiers remain private.

04

Private workload contents are not published through public pages.

05

Network guardrails control supported execution paths.

Two ways to participate

Use the network.
Or help build it.

Run a workload → Become a Host →