Early access
Early access · API

Price your workload, then create your account.

Set your volumes and see our indicative early-access rates next to the public list prices of frontier APIs. An administrator reviews every account and grants access.

Workload

Use case

Pick a starting point, then adjust the sliders.

Text volume
10,000
1,000
Prompt, context and documents.
200
Audio volume
0
Used when Clio is selected.
0
Used when Calliope is selected.

Models

Our models

Indicative early-access rates. Sizes and measurements are on the model cards.

Compare with

Public list prices, ordered from cheapest to most expensive.

Claude Opus 5.5
OpenAI gpt-transcribe
ElevenLabs v4

Create account

Create your early-access account. An administrator reviews every account and grants access; you then receive your endpoint and API key by email. Your calculator settings are sent with the request.

Unit prices

The rates behind the calculator

Text models, € per 1M tokens
ModelProviderInputOutput
Iris 8BHellas Inference€0.068€0.27
Metis 12BHellas Inference€0.35€1.40
Gemini 3.8 FlashGoogle€0.68€3.39
Claude Sonnet 5.5Anthropic€1.81€9.05
Gemini 4 CarbonGoogle€1.81€9.05
Claude Opus 5.5Anthropic€3.62€18.10
GPT-6 AstraOpenAI€9.05€45.24
Speech to text, € per audio minute
ModelProviderPer minute
Clio 1.5BHellas Inference€0.0091
Scribe v3ElevenLabs€0.0030
gpt-transcribeOpenAI€0.0041
Nova-3 MultilingualDeepgram€0.0053
Voice, € per minute of speech
ModelProviderPer minute
Calliope 3BHellas Inference€0.00069
tts-1-hdOpenAI€0.0244
Aura-2Deepgram€0.0244
v4 TurboElevenLabs€0.0326
v4ElevenLabs€0.0651
  1. Frontier prices are public list prices in US dollars, standard tier, read from each provider's pricing page on 15 November 2026 and converted at the ECB reference rate of 13 November 2026 (€1 = $1.1052). They change often; check the provider before you rely on them.
  2. Our rates are indicative early-access rates, derived from throughput we measured on one NVIDIA H200 (measurements). They are not a quotation and may change before general availability.
  3. Voice priced per character is converted at 900 characters per minute of speech. No batch, caching or volume discounts are applied on either side. Taxes are excluded. A month is 30 days.
  4. The models are not the same. Frontier models are large general-purpose models; ours are smaller open-weight models tuned for Greek business tasks. Price is one side of the comparison: measure quality on your own task before you choose.