Pricing
Run open models on
infrastructure you own.
We install our inference system inside your cloud, so you can power agents efficiently with less hardware and better performance. Deploy in your cloud, run it on the edge, or trial it first on our hosted API.
For enterprises
Half the GPUs, faster throughput, better capability, better economics
Our serving infrastructure runs more concurrent agents on the same hardware, so a node does the work that used to take two. You keep the savings, we charge a fraction of it, with hands-on setup by our team of world-class AI researchers.
50%
Same work with half the GPUs
3.5x
Faster token throughput
10x
Longer usable context
Expert
Support from our team
If you have GPUs
We install in your cloud
Already running GPUs? We deploy our inference system directly onto the fleet you have today, and your team is serving open models in your own cloud within a week.
If you need GPUs
We get you the hardware first
No spare GPUs? We connect you to our compute partners or work with your cloud provider of choice, secure the lowest cost per GPU we can, then install and run the system for you.
Estimate your savings
GPU price, per hour
Deployment size, GPUs
Generic serving
$186,880/mo
With Subconscious
$116,800/mo
50% the GPUs + our per GPU pricing*
You save
$70,080/mo
38% lower per month
* An estimate of our cost per node using our inference system, which covers everything we provide: our inference system, a routing gateway and supporting infrastructure, a frontend for API key management and usage monitoring, and dedicated setup and support from our team of world-class AI researchers.
On the edge
Run agents on devices that couldn't before
Our runtime fits capable agents into the limited memory of laptops, workstations, and edge hardware, unlocking on-device workloads that generic serving cannot run. Licensed per device.
For device makers and teams
Agents that run fully on-device
Ship capable agents that run offline, keep data on the device, and work where there is no connection. Priced per device, on the hardware you already own.
Talk to Us About On Device Deployments →Run via API
Use our managed inference service
To experience the power of our system without deploying on your own hardware, run open models on our managed inference service to see the speed and quality for yourself. We'll double your first top-up, up to $100.
| Model | Cached tokens | Input tokens | Output tokens |
|---|---|---|---|
| GLM-5.2 Open-source frontier coding model subconscious/glm-5.2 | $0.26 | $1.40 | $4.40 |
| TIM-Qwen3.6 27B Multimodal post-trained model subconscious/tim-qwen3.6-27b | $0.15 | $0.30 | $3.00 |
Prices in USD per 1M tokens. With efficient caching, we see upwards of a 95% cache hit rate on agentic coding workloads, so most input tokens bill at the cached rate.
Frontier closed models, for comparison
| Model | Cached tokens | Input tokens | Output tokens |
|---|---|---|---|
| Claude Opus 4.8 | $0.501.9x | $5.003.6x | $25.005.7x |
| GPT-5.5 | $0.501.9x | $5.003.6x | $30.006.8x |
| Gemini 3.1 Pro | $0.200.8x | $2.001.4x | $12.002.7x |
Published list prices in USD per 1M tokens. The factor under each price is how much more expensive it is than our GLM-5.2 rate above.