Infrastructure Update

Fresco Is Now on Dedicated Servers

Sennoric moved off shared Hugging Face Spaces onto dedicated hardware under full control. The result: responses in seconds instead of minutes, no rate-limit contention, and infrastructure that scales with demand.

Sennoric · July 20, 2026
~80×
Faster responses
220s → 2–3s
0
Shared tenants
no rate-limit contention
100%
Uptime ownership
full stack control
What changed

Why Sennoric moved

Fresco started on a shared Hugging Face Space: easy to deploy, and good enough to get the model into people's hands quickly. But shared hosting has hard limits that started creating problems.

The problem with shared hosting

Hugging Face Spaces run on shared infrastructure. When demand spikes — because someone else's model goes viral or multiple apps share the same GPU allocation — Fresco's response times climb and requests get throttled. Hardware, scheduling, and networking are outside Sennoric's control.

For a lab building its own models, that's a ceiling with no workaround. Dedicated infrastructure was needed: dedicated GPUs, no noisy neighbors, and the ability to scale independently.

Before — Shared HF Space
×Shared GPU with other models
×Rate limits not configurable
×Cold starts after idle periods
×No control over scheduling or priority
×Dependent on HF infrastructure uptime
After — Dedicated Servers
Dedicated GPU running Fresco exclusively
Configurable rate limits under full control
Always warm — no cold starts
Full scheduling control over priority and throughput
End-to-end ownership of the entire serving stack
What this means for you

Faster responses. Fresco now runs on dedicated hardware with no contention from other workloads. Prompts that took around 220 seconds on the shared Space now complete in 2–3 seconds.

More reliable. No more intermittent slowdowns when HF infrastructure is under load. Sennoric owns the uptime end to end.

Same endpoint. The API hasn't changed. api.sennoric.com still serves the same OpenAI-compatible /v1/chat/completions endpoint. No SDK changes, no config changes.

The model hasn't changed. This is the same Fresco 1.2.5: same weights, same safety alignment, same benchmarks. The upgrade is purely infrastructure. Think of it as moving from a rented apartment to a house Sennoric built.
How it happened

The migration timeline

Moving off shared hosting while keeping the model available wasn't possible overnight. Here's how it unfolded.

July 6, 2026
Better HF hardware
Upgraded to a larger Hugging Face Space with more GPU allocation. A short-term fix that improved throughput but didn't solve the shared-hosting problem.
July 12, 2026
Migration begins
Fresco and Glyph began migrating onto dedicated hardware. Both models were temporarily unavailable or unstable during the transition.
July 20, 2026
Migration complete
Fresco is fully live on dedicated infrastructure. Faster, more reliable, and under Sennoric's complete control. Glyph is also being migrated and will be available again shortly.
What's next

This migration is step one. Dedicated servers give Sennoric the foundation to run more experiments, serve more users, and iterate faster on Fresco 1.3. With full control over the infrastructure, Sennoric can now benchmark under realistic conditions, test scaling strategies, and push the model harder without hitting shared limits.