🇬🇧UK Sovereign AI
Frontier AI Models
on UK Infrastructure
Argyll Data Development provides secure, UK-hosted inference and hands-on AI enablement – helping organisations deploy Sovereign AI across internal workflows and customer-facing products while staying in control.
Frontier AI should not mean foreign jurisdiction.
Most AI infrastructure decisions force a tradeoff: capability on one side, operational control on the other. We're built for teams that need both.
We give UK enterprises a sovereign path to production AI: OpenAI-compatible API access to advanced models, UK-hosted infrastructure for sensitive data, and the control teams need to scale with confidence.
- Stay in control – with no surprise model deprecations.
- Keep sensitive data on British soil.
- Avoid CLOUD Act exposure and foreign jurisdiction risk.
Built for UK data residency, GDPR, the UK Data Act, and sector-specific requirements across finance, healthcare, legal, government, and defence.
Purpose-built AI inference on SambaRack™ SN40Ls
Trusted by SambaNova as their official UK partner, we run your AI workloads on SambaRack™ SN40Ls purpose-built for production inference. Their dataflow architecture is designed to reduce unnecessary data movement by keeping more computation and intermediate data local to the chip, helping lower memory bottlenecks resulting in higher throughput, lower latency, and better energy efficiency than conventional GPU-based approaches for the workloads they are optimized to accelerate.
Benchmark snapshot prepared May 27, 2026. Public speeds vary by prompt length, reasoning mode, batching, region, and provider load.
The Model
Powered today by MiniMax M2.7
The model powering Argyll's infrastructure today is MiniMax M2.7 – an open-weight frontier model built specifically for agentic workloads: autonomous multi-step tasks, tool use, and long-running agent loops, not just single-turn responses. Because it's open-weight, we can run M2.7 entirely on UK infrastructure, with no licensing tie to a US provider, ensuring your company gets powerful, agent-ready AI under your control, with data and computation fully within UK jurisdiction and no dependency on US-based model APIs.
MiniMax M2.7 is a 230B-parameter sparse mixture-of-experts (MoE) model. Independent analysis from Artificial Analysis confirms M2.7 delivers strong performance on real-world agentic tasks with reduced hallucinations, achieving a GDPval-AA Elo of 1,494.
What that means in production
- Built for agents – multi-step tool use, autonomous task execution, and sustained agent loops, with a 97% skill-adherence rate on complex multi-step tasks
- 56.2% on SWE-Pro – one of the toughest real-world software engineering benchmarks, covering bug troubleshooting, code security review, and log analysis
- 78% on SWE-bench Verified – validating real-world software engineering capability across diverse repositories
- 205k context window – entire contracts, codebases, or knowledge bases in a single call
- High-throughput inference on SN40L – sustained performance for real-time agentic applications on SambaRack™ SN40Ls
As the open-weight landscape develops, Argyll's infrastructure will support more models. The sovereign guarantee stays the same; the roster grows.
How it works
One path from pilot to production.
Start with MiniMax on UK-hosted infrastructure, test through a familiar API, then scale into dedicated capacity, hands-on enablement, and sovereign production operations when you're ready.
Choose MiniMax
Start with MiniMax for fast agent execution, summarisation, reasoning, and secure business workflows.
Change the endpoint
Use OpenAI-compatible chat completion patterns so engineering teams can evaluate quickly without rebuilding existing workflows.
Scale with control
Move from testing to production with UK-hosted inference, reserved capacity, hands-on enablement, and enterprise deployment support.
curl -H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"stream": true,
"model": "MiniMax-M2.7",
"messages": [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": "Hello!"
}
]
}' \
-X POST https://api.argylldev.ai/v1/chat/completionsfrom sambanova import SambaNova
client = SambaNova(
api_key="",
base_url="https://api.argylldev.ai/v1",
)
response = client.chat.completions.create(
model="MiniMax-M2.7",
messages=[{"role":"system","content":"You are a helpful assistant"},{"role":"user","content":"Hello!"}],
temperature=0.1,
top_p=0.1
)
print(response.choices[0].message.content) import gradio as gr
gr.load("MiniMax-M2.7", accept_token=True).launch()Pricing
Simple, scalable plans.
Start testing with free tokens, move into production on demand, or reserve capacity for high-volume workloads.
Free
Explore sovereign AI inference without a credit card.
- 500,000 tokens
- Access to MiniMax M2.7
- No credit card required
- Restricted rate limits
- Community support
Developer
Pay as you go for production inference that scales on demand.
- Pay-per-token pricing
- Preview model access
- Standard rate limits
- Standard support
Enterprise
Scale production traffic with volume pricing, dedicated support, and capacity options.
- Volume-based pricing
- Access to all production models
- Dedicated infrastructure options (including reserved capacity)
- Custom rate limits
- Priority support and SLA
What does sovereign AI inference cost you?
Example workloads
Ready to build on UK-hosted AI infrastructure?
Start with free credits, validate the API, then move into production or reserve dedicated capacity when your workload is ready.