Models & pricing

GPUniq serves 370+ models — text, image, video, music, speech and restoration — behind one API key and one USD balance. The catalog changes weekly, so this page documents the rules. For the numbers, ask the API:

curl https://api.gpuniq.com/v1/llm/models/catalog

models = client.llm.models()
for model in models["models"]:
    print(model["slug"], model["retail_input_usd_per_mtok"])

curl https://api.gpuniq.com/v1/openai/models \
  -H "Authorization: Bearer gpuniq_your_key"

GET /v1/llm/models/catalog is public and needs no key. It is the only authoritative price list — the model cards in /chat read from it too.

What the catalog returns

One entry per model. The fields that matter when you are choosing or budgeting:

The response also carries default_model — currently claude-haiku-4-5, used when a chat request omits model.

Text & reasoning models

The main vendor families. Exact slugs and prices come from the catalog; this is the map of what's on the shelf.

Filter the catalog by category to narrow it — flagship for the top of each line, fast for the cheap high-throughput tier.

How pricing works

Every surface bills against the same USD balance, but the unit differs:

A failed, refused or timed-out media job costs nothing. For media, the kickoff response carries estimated_cost_usd and the completion response carries the authoritative cost_usd and your new balance_usd.

Discounts

Most models are sold below vendor list price, but the discount is per model, not a flat platform rate — it ranges from 7.5% to 30% depending on what the weights cost us wholesale. Read discount_percent_label for the public discount on a given slug and discounted for whether one applies at all.

Account-level discounts stack on top and appear in the same record as user_discount_percent / user_tier_discounts, so the catalog you fetch with your key already reflects your price.

The open-weight tier

Open-weight models — among them deepseek-v4-pro, deepseek-v4-flash, deepseek-v3.2, gpt-oss-120b, GLM-5.x, Kimi K2.x, MiniMax M3 and Gemma — are not priced from a fixed catalog number. They are priced from the cheapest live rate available for those weights, plus our margin, and that moves as the market moves. Proprietary lines such as Qwen Plus / Max / Flash are not in this tier and carry a fixed catalog price.

This is why no number for them is printed here: anything published would be stale within days. Read the live figure from the catalog or the model card in /chat.

Your balance

Requests are billed directly against your GPUniq account balance in USD. There is no separate token pool — prepaid token packages and ruble-to-token conversion were retired, and the balance is shared with every other GPUniq surface, so one deposit covers GPU rentals, volumes and every model here.

balance = client.llm.balance()
print(f"Available: ${balance['balance_usd']:.4f} USD")

curl https://api.gpuniq.com/v1/llm/balance \
  -H "Authorization: Bearer gpuniq_your_key"

Top up from the web dashboard → Balance: card or SBP through T-Bank, or card through Stripe.

Usage history

Per-request detail with prompt / completion / cached / reasoning tokens and the USD charged.

history = client.llm.usage_history(limit=50, offset=0)
for log in history["logs"]:
    print(f"{log['model']}: {log['total_tokens']} tokens — ${log['cost_usd']:.6f}")

Model throughput

GET /v1/llm/models/{slug}/performance returns measured throughput for one model — useful when two models are close on quality and you are choosing on latency. The same figure appears in the catalog as tokens_per_sec, with tokens_per_sec_approx flagging whether it is a measurement or an estimate.

curl https://api.gpuniq.com/v1/llm/models/claude-opus-5/performance \
  -H "X-API-Key: gpuniq_your_key"