Run AI agents, subagents and continuous workloads without paying for every token. Choose how many Workers can operate in parallel.
Premium models think. FlatCompute models do the work. We take high-quality open models, run them in specialized zones, and give you unlimited tokens through Active Workers and managed queues — for one fixed monthly price.
REASONING & QUALITY
QL-EU-001
Dense models are slower and more expensive, but maintain better attention, lose focus less often, and excel at detail-oriented work. Best for complexity, coding, debugging and technical analysis.
AGENTS & AUTOMATION
WH-EU-001
High-volume inference for agents and continuous workloads. MoE models are faster and more efficient, but less precise on details — they lose attention more easily. Best for throughput and cost-efficiency.
Target: 30 t/s average. Not a cap — if the node has spare compute, you get everything it can give.
VISION + TEXT
MM-EU-001
Multimodal models excel at visual tasks — screenshots, interfaces, documents and images. They are MoE, making them versatile for general-purpose workloads where vision is essential.
Target: 30 t/s average. Not a cap — if the node has spare compute, you get everything it can give.
Current model: Qwen3.6-35B-A3B
| FEATURE | Starter | Builder | Pro |
|---|---|---|---|
| Active Workers | 1 | 2 | 4 |
| Context window | 64K | 128K | 256K |
| RPM | 20 | 60 | 200 |
| API keys | 1 | 3 | 5 |
| Price / mo | 19.90€ · Founding 1.99€ | 39.90€ | 79.90€ |
| Comparative buying tokens Mtok/€ | |||
| Tokens Mtok/€ (30t/s · 5:1) | ≈467Mtok · ≈100€ | ≈933Mtok · ≈200€ | ≈1866Mtok · ≈401€ |
≈78M output tokens/mo at 30 t/s sustained (1 Worker). OpenRouter avg for Qwen 3.6 35B A3B: $0.10/M input, $0.90/M output (≈€0.09/M in, €0.83/M out). At 5:1–10:1 input:output ratio, comparable API cost ≈€100–€135/mo vs our launch price €1.99 first month, then €19.90/mo (Starter). Ends September 1st. No per-token billing — the price stays the same no matter how much you generate.
Current model: Qwen3.8-27B
Founding: 50% off first month
| FEATURE | Starter | Builder | Pro |
|---|---|---|---|
| Active Workers | 1 | 2 | 4 |
| Context window | 64K | 128K | 256K |
| RPM | 20 | 60 | 200 |
| API keys | 1 | 3 | 5 |
| Price / mo | 59.90€ | 119.90€ | 239.90€ |
| Comparative buying tokens Mtok/€ | |||
| Tokens Mtok/€ (30t/s · 5:1) | ≈467Mtok · ≈322€ | ≈933Mtok · ≈644€ | ≈1866Mtok · ≈1288€ |
≈78M output tokens/mo per Worker at 30 t/s sustained. OpenRouter avg for Qwen3.8-27B: $0.35/M input, $2.75/M output (≈€0.32/M in, €2.53/M out). At 5:1–10:1 ratio, comparable API cost ≈€318–€439/mo vs our flat €59.90/mo (Starter). No per-token billing — the price stays the same.
Current model: Gemma4 26B-A4B
Founding: 50% off first month
| FEATURE | Starter | Builder | Pro |
|---|---|---|---|
| Active Workers | 1 | 2 | 4 |
| Context window | 64K | 128K | 256K |
| RPM | 20 | 60 | 200 |
| API keys | 1 | 3 | 5 |
| Price / mo | 19.90€ · Founding 9.95€ | 39.90€ · Founding 19.95€ | 79.90€ · Founding 39.95€ |
| Comparative buying tokens Mtok/€ | |||
| Tokens Mtok/€ (30t/s · 5:1) | ≈467Mtok · ≈49€ | ≈933Mtok · ≈99€ | ≈1866Mtok · ≈197€ |
≈78M output tokens/mo per Worker at 30 t/s sustained. OpenRouter avg for Gemma4 26B-A4B: $0.07/M input, $0.34/M output (≈€0.06/M in, €0.31/M out). At 5:1–10:1 ratio, comparable API cost ≈€49–€74/mo vs our flat €19.90/mo (Starter). Multimodal value goes beyond text-token pricing. No per-token billing — the price stays the same.
How we estimate savings: Output tokens/mo = 30 t/s × 3,600s × 24h × 30 days × Active Workers. For Workhorse EU-1 (Qwen 3.6 35B A3B), OpenRouter average pricing is $0.10/M input, $0.90/M output (≈€0.09/M in, €0.83/M out at 0.92 EUR/USD). Agent workloads typically have 5:1–10:1 input-to-output token ratios. Example: Starter (1 Worker) at 30 t/s → ≈78M output tokens (≈€65) + 389M–778M input tokens (≈€35–€70) → Total ≈€100–€135/mo on a per-token API. Our flat price: €19.90/mo.
The 30 t/s is our design target, not a cap — if the node has spare compute, you get everything it can give. Actual throughput varies with context length, workload, and zone utilization. Token prices sourced from OpenRouter (Aug 2026) and may change. Not a guarantee of savings.
A Worker is one active AI generation. When all your Workers are busy, additional tasks wait in your queue. More Workers means more parallel work; fewer Workers means tasks queue up but still complete.
One Worker active. Extra tasks queue and run one at a time.
Three Workers, two active. Tasks run in parallel; overflow queues.
Four Workers, three active. Queue empty — agents run at full parallel.
Unlimited tokens means no monthly token allowance and no per-token billing. Your agents generate as many tokens as they need — the price stays the same every month.
No monthly token allowance and no per-token billing. Your agents generate as much as they need — the price stays the same.
A Worker is one active AI generation. When all your Workers are busy, additional tasks wait in your queue — they are never dropped.
Tasks that can't run immediately queue up and complete in order. Higher-tier plans get priority queueing so your work jumps ahead.
Point your existing client at our endpoint, change one URL, ship. Works with LangChain, LlamaIndex, the OpenAI SDK, and anything that speaks OpenAI.
Each zone is tuned for a workload: Workhorse for agents, Quality for reasoning, Multimodal for vision + text. Pick what your agents do.
One price, every month. No token counting, no budgeting, no surprise bills. Predictable like rent.
Reserve a seat before a zone opens. Founding members are locked in for the life of the zone — no migration, no price changes.
Need more parallel work? Upgrade your plan or move to a bigger zone. Your API key, history, and settings follow you.
Founding memberships are limited. Once a zone fills, it fills — we do not oversell. Reserve early to lock in your membership for the life of the zone.
No monthly token allowance and no per-token billing. Your agents generate as many tokens as they need — the price stays the same every month. We manage capacity through Active Workers and queues, not token caps.
Start with Workhorse EU-1. No token counting, no per-token billing, no surprise bills. Your agents run as much as they need.
Start with Workhorse →