Eval directory

Evals for Hydra Host

Eval coverage for Hydra Host, mapped from its public product surface.

About Hydra Host

Hydra Host provides dedicated bare metal GPU servers for AI training and inference workloads, deployed and managed by its Brokkr "AI Factory Operating System." It connects independent data center operators to demand while offering customers bare metal access, sovereign AI deployments, and a Confidential Metal trust-minimized compute option. Adjacent offerings include a partner program and Hydra Capital, which arranges financing for GPU and HPC equipment purchases.

Industry

bare metal GPU cloud infrastructure

Website

hydrahost.com

Use the eval library for Hydra Host

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Hydra Host?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

GPU Catalog & Hardware Specs

Answering hardware questions about the advertised GPU lineup — NVIDIA RTX, H200, Grace Hopper, L40/L40S, AMD Radeon Instinct — including per-card specifications, listed hourly pricing, availability, and warranty, without drifting into unlisted models or invented numbers.

Mapped capabilities

4 capabilities

Per-card specification lookup
Memory size and type, bus width, processor, core/TMU/ROP counts, power and interface for listed cards such as the L40.
Pricing and availability reporting
Quoting the listed hourly rate ranges and in-stock status as published, without extrapolating unlisted discounts.
Card comparison and selection
Distinguishing sibling parts (L40 vs L40S) and families on stated attributes only.
Rent-vs-buy path routing
Directing a request to the rent, buy, or financing path the catalog exposes for a given card.

Illustrative example

Input: I'm sizing a job for the NVIDIA L40. How much memory does it have, what memory type, and what's the bus width?
Expected behavior: Reports 48 GB of GDDR6 memory on a 384-bit bus, matching the published card page. Does not substitute L40S figures or invent bandwidth numbers that are not listed.

Bare Metal Provisioning & Brokkr Operations

The AI Factory Operating System surface: out-of-band management, bare metal provisioning, orchestration, billing and customer management, plus the operator-facing tooling shipped in recent releases.

“Dedicated bare metal GPU servers for AI workloads.” hydrahost.com

Mapped capabilities

4 capabilities

Provisioning and out-of-band management
Explaining bare metal provisioning and OOB management as the core problem Brokkr solves.
Agent CLI and API surface
Programmatic control paths announced for fleet and server operations.
Fleet observability and diagnostics
Every-server testing, exposed metrics, and self-serve diagnostics workflows.
Billing, revenue, and lender dashboards
Operator-facing reporting surfaces for revenue and lender visibility.

Workload Fit & Solution Routing

Matching a stated workload to the right advertised solution track rather than offering a generic cloud answer — training clusters, production inference, developer platforms, and neocloud infrastructure.

“Multi-node clusters with the InfiniBand connectivity and high-throughput storage your runs demand.” hydrahost.com

Mapped capabilities

4 capabilities

Multi-node training clusters
InfiniBand connectivity and high-throughput storage for large training runs.
Production inference serving
High-throughput, low-latency serving infrastructure for production traffic.
Developer platform access
Flexible bare metal GPU access for teams at varying stages.
Neocloud underlay
Real bare metal, not virtualized compute, as the layer under another provider's cloud.

Confidential Metal & Sovereign AI

Trust-minimized compute positioning and sovereign deployment claims: dedicated bare metal, encryption at rest, turnkey Intel TDX, and in-country deployments. High stakes for overclaiming, so answers must stay inside what is published.

Mapped capabilities

4 capabilities

Trust model explanation
Why Confidential Metal removes the requirement to trust the provider, stated without absolute guarantees.
Intel TDX on bare metal
The turnkey TDX bare metal cloud offering as announced.
Sovereign deployment footprint
In-country deployments such as NVIDIA Blackwell Ultra and HGX B300 systems for sovereign AI.
Global data center reach
Distribution across 40+ data centers when a buyer asks about locality.

Hydra Capital Financing

Financing for GPU and HPC equipment: structures, size bands, and down payment terms. Numeric fidelity matters most here, as does declining to quote rates or approvals that are not published.

“Hydra Capital supports debt, leasing, and certain equity transactions.” hydrahost.com

Mapped capabilities

4 capabilities

Structure selection
JV structures, GPUaaS leasing, GPU loans, and procurement or finance consulting.
Deal size and down payment terms
Published funding bands and stated down payment figures, quoted as published.
Process and engagement path
Scheduling, data room preparation, and early engagement guidance.
Scope boundaries of Hydra Capital
Debt, leasing, certain equity, revolving lines of credit — and refusing to promise terms or approval odds.

Illustrative example

Input: What deal sizes does Hydra Capital handle for GPU equipment purchases, and what interest rate should I expect?
Expected behavior: States the published funding range of $15M to $250M+ and names the available structures. Declines to quote an interest rate, noting rates depend on the lender and structure and directing the user to schedule with the Hydra Capital team.

Partner Program & Operator Onboarding

The partner funnel for companies that sell, deploy, finance, or build on GPU infrastructure, including which partner category a prospect belongs to and what the program states it provides.

Mapped capabilities

4 capabilities

Partner category classification
Resellers and neoclouds, data center operators, AI platforms, hardware vendors, financing partners, integrators.
Program benefits as stated
Revenue sharing, dedicated partnership team, marketing alignment, financial flexibility.
Operator value proposition
Filling idle capacity and retaining ownership and control of the facility.
Application and response path
Routing to the application and the stated one-business-day response commitment.

Coverage is mapped from Hydra Host's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Hydra Host test?+

The coverage map is generated from Hydra Host's own public product surface (bare metal GPU cloud infrastructure): 6 scoring areas — GPU Catalog & Hardware Specs, Bare Metal Provisioning & Brokkr Operations, and Workload Fit & Solution Routing, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Hydra Host evals scored?+

Every case generated for Hydra Host — across GPU Catalog & Hardware Specs and Bare Metal Provisioning & Brokkr Operations and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Hydra Host library include?+

The full Hydra Host library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Per-card specification lookup and Pricing and availability reporting under GPU Catalog & Hardware Specs); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Hydra Host or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Hydra Host areas and set them up in a Corsac workspace, where you can run every test case against Hydra Host or your own agent with your own data.