Inference provider · Miami, Florida
Serious capacity for models you can’t get anywhere else.
Lore Hex runs a high-throughput inference platform out of the United States. We serve our own model line alongside frontier open models, on one OpenAI-compatible API, under one data policy: we do not train on your traffic and we do not keep it.
The model line
Three of our models are built in house. They are designed around workloads where a single general-purpose model tends to underperform: long-horizon research, provenance- constrained deployments, and security work.
Prometheus 2.0
lorehex/prometheus-2.0
A research model with a one-million-token context. Built for multi-step investigation across large document sets — diligence packets, literature sweeps, competitive teardowns — where the work is holding many sources in view at once and synthesizing across them.
Liberty 2.0
lorehex/liberty-2.0
A general-purpose model served exclusively on US-origin open weights, running on US infrastructure. For buyers whose procurement or regulatory posture constrains where their model weights and their inference come from.
OpenPatcher-G1
lorehex/openpatcher-g1
A security model for vulnerability triage and remediation: reading a large codebase, localizing a defect from an advisory or a crash, and drafting a minimal patch with a regression test that proves it.
Archimedes
lorehex/archimedes-small-1.0, lorehex/archimedes-large-1.0
Two tiers designed for government and national security applications. Small is priced for high-volume classification, extraction, and routing where cost per call dominates; large handles synthesis and agentic work across big document sets.
Frontier open models
lorehex/glm-5.3-flash, lorehex/kimi-k3
We also serve widely-used open models on our own capacity, so a workload can move between our line and the open ecosystem without changing clients, keys, or data posture.
| Model | Context | Built for |
|---|---|---|
| lorehex/prometheus-2.0Prometheus 2.0 | 1M | Deep research |
| lorehex/liberty-2.0Liberty 2.0 | 256K | US-origin weight provenance |
| lorehex/openpatcher-g1OpenPatcher-G1 | 1M | Vulnerability triage |
| lorehex/archimedes-large-1.0Archimedes-large 1.0 | 256K | Government & national security |
| lorehex/archimedes-small-1.0Archimedes-small 1.0 | 128K | Government & national security |
| lorehex/glm-5.3-flashGLM 5.3 Flash | 1M | Long-horizon agents |
| lorehex/kimi-k3Kimi K3 | 1M | Agentic tool use |
Full catalog and API usage →
· /api/v1/models
How we handle traffic
We do not train on prompts or completions, and we do not retain them. Request and response bodies live in memory for the duration of the call and are never written to durable storage. What we keep is billing metadata — timestamp, model, token counts, latency, status — with no prompt or completion content in it.
The full commitment, including retention windows and what happens under legal process, is in the data policy.
Contact
- Partnerships and provider onboarding
- providers@lorehex.co
- Support
- support@lorehex.co
- Security disclosure
- security@lorehex.co
- Entity
- Lore Hex, Inc. · Miami, Florida, United States