MODEL CATALOGUE
The engine behind every AREVIA service. Each model carries the name of an Armenian landmark and is built to global standards: pinned versions, published evaluations and one API for all of them.
TEXT
Generally availableԱրագածАрагац
Flagship reasoning model
Our most capable model for complex analysis, multi-step reasoning, agents and long documents. Designed for tasks where a wrong answer costs more than a slow one.
aragats-2TEXT
Generally availableՍևանСеван
Balanced general-purpose model
The workhorse behind our assistants, order-taking and document workflows. Strong quality at a fraction of the flagship’s cost.
sevan-2TEXT
Generally availableԴիլիջանДилижан
Compact low-latency model
Small enough for on-premises hardware and air-gapped sites, fast enough for real-time classification, routing and extraction at scale.
dilijan-1.5EMBEDDINGS
Generally availableԳառնիГарни
Embeddings for search
Dense vectors for search, clustering and retrieval across languages, so a question asked in one language finds the answer written in another.
garni-embed-1SPEECH
BetaՀրազդանРаздан
Speech recognition and synthesis
Streaming transcription and natural voice synthesis for contact centres and voice services, including calls that switch between languages.
hrazdan-s1SAFETY
Generally availableԷրեբունիЭребуни
Safety and data-protection model
Detects and masks personal data, flags prompt-injection attempts and filters unsafe content before it reaches your users or your logs.
erebuni-guard-1| MODEL | TYPE | API IDENTIFIER | CONTEXT WINDOW | INPUT | OUTPUT | TYPICAL LATENCY | STAGE |
|---|---|---|---|---|---|---|---|
| Aragats | Text | aragats-2 |
128K | $2.00 | $8.00 | ≈ 900 ms to first token | Generally available |
| Sevan | Text | sevan-2 |
128K | $0.40 | $1.60 | ≈ 350 ms to first token | Generally available |
| Dilijan | Text | dilijan-1.5 |
32K | $0.08 | $0.30 | ≈ 120 ms to first token | Generally available |
| Garni | Embeddings | garni-embed-1 |
8K | $0.02 | n/a | ≈ 40 ms per batch of 32 | Generally available |
| Hrazdan | Speech | hrazdan-s1 |
n/a | $0.006 per minute of audio | Real-time streaming | Beta | |
| Erebuni | Safety | erebuni-guard-1 |
32K | $0.10 | n/a | ≈ 30 ms per request | Generally available |
Prices in US dollars per one million tokens, excluding VAT. Latency is the median measured in Yerevan on the Cloud API.
Every model has a dated, immutable version identifier. Your integration never changes behaviour without your decision.
We announce deprecations at least twelve months in advance and support your migration testing.
Every release comes with evaluation results on public and industry benchmarks.
NEXT STEP
Tell us about your process. Within five working days we prepare a pilot plan: scope, success metrics, timeline and a fixed price.