Skip to content
siema_ai

Models and pricing

You call models by alias. An alias is a stable name that we can point at a newer or better-quantised model without you changing code. GET /v1/models tells you what each alias can do and what it costs right now.

Alias Underlying model (V1) Capabilities Context Max output Use it for
polish-fast Bielik-Minitron 7B v3 chat 8 192 2 048 Classification, extraction, rewriting, FAQ answers, high-volume Polish text.
polish-pro Bielik 11B v3 chat 8 192 2 048 The default for Polish. Longer conversations, business and legal text.
advanced Qwen 27B chat 32 768 4 096 Harder reasoning, code, and multilingual tasks. Costs more per token.
embed BGE-M3 embeddings 8 192 0 Search, clustering and retrieval. 1 024-dimensional vectors, 100+ languages.

No V1 alias has the vision capability yet, so none of them accepts image input today — see Chat completions for what happens if you send one anyway.

Context and output limits above are provisional until the launch benchmark confirms them. Read context_length and max_output_tokens from the API rather than hard-coding these numbers.

Prices

Section titled “Prices”

Prices are per one million tokens, in PLN, and are returned by the API in micro-PLN (1 PLN = 1 000 000 micro-PLN) so that arithmetic is exact:

Terminal window
curl -s https://api.siema-ai.pl/v1/models -H "Authorization: Bearer $SIEMA_API_KEY" \
  | python3 -c '
import json, sys
for m in json.load(sys.stdin)["data"]:
    mid = m["id"]
    p = m["pricing"]
    pin = p["input_micro_pln_per_1m"] / 1e6
    pout = p["output_micro_pln_per_1m"] / 1e6
    print(f"{mid:12} in {pin:7.2f} PLN/M  out {pout:7.2f} PLN/M")
'

Launch prices will be published here and in the changelog. Until then, the prices returned by GET /v1/models — and quoted anywhere in these docs — are placeholders used for testing, not a price list:

Alias Input (PLN / 1M tokens) Output (PLN / 1M tokens)
polish-fast 0.80 1.60
polish-pro 1.50 3.00
advanced 4.00 8.00
embed 0.20 — (embeddings have no output tokens)

Input and output tokens are priced separately; embeddings bill input tokens only. Price changes take effect from a published date and never apply to requests already made. See Billing for how a request’s cost is computed.

Which alias?

Section titled “Which alias?”
  • Start with polish-pro. If quality is fine and you need lower cost or latency, try polish-fast on the same prompts.
  • Use advanced when the task involves harder reasoning, code, or languages other than Polish and English.
  • Use embed for anything retrieval-shaped. Pair it with polish-pro for Polish RAG in your own stack (siema_ai does not host documents).

Model changes

Section titled “Model changes”

When we change the model behind an alias we announce it in the changelog at least 14 days ahead, run the new model on a share of traffic first, and keep the alias’s context and output limits the same or larger. The model field in responses always echoes the alias you asked for.