Concepts

Pricing

Per-token USDC pricing — the paid model catalog plus a free NVIDIA NIM tier. No platform fee on the hosted gateway.

Solvela charges per token based on the upstream provider's list price. The gateway software includes a configurable platform fee (default 5%); the hosted gateway at api.solvela.ai runs with the fee suspended, so you pay only the provider cost. All prices are in USDC. There are no monthly fees, no minimum spend, and no subscription.

Fee structure

On the hosted gateway, the fee is suspended, so:

total = provider_cost

On a self-hosted gateway with the default fee enabled:

total = provider_cost * 1.05

Every 402 response includes a cost_breakdown showing the split. On the hosted gateway (fee suspended):

{
  "cost_breakdown": {
    "provider_cost": "0.002500",
    "platform_fee": "0.000000",
    "total": "0.002500",
    "currency": "USDC",
    "fee_percent": 0
  }
}

On a self-hosted gateway with the fee enabled (default 5%):

{
  "cost_breakdown": {
    "provider_cost": "0.002500",
    "platform_fee": "0.000125",
    "total": "0.002625",
    "currency": "USDC",
    "fee_percent": 5
  }
}

Amounts in the accepts[].amount field are in atomic USDC units (6 decimal places). 2625 = $0.002625 USDC.

Model pricing table

All prices are USDC per million tokens (provider cost only; the hosted gateway has no platform fee).

OpenAI

ModelModel IDInput ($/M)Output ($/M)Context
GPT-5.2openai/gpt-5.2$1.75$14.00400K
GPT-4oopenai/gpt-4o$2.50$10.00128K
GPT-4o Miniopenai/gpt-4o-mini$0.15$0.60128K
o3openai/o3$2.00$8.00200K
o3 Miniopenai/o3-mini$1.10$4.40200K
o4 Miniopenai/o4-mini$1.10$4.40200K
GPT-4.1openai/gpt-4.1$2.00$8.001M
GPT-4.1 Miniopenai/gpt-4.1-mini$0.40$1.601M
GPT-4.1 Nanoopenai/gpt-4.1-nano$0.10$0.401M
GPT-OSS 120Bopenai/gpt-oss-120b$0.00$0.00128K

Anthropic

ModelModel IDInput ($/M)Output ($/M)Context
Claude Opus 4.8anthropic/claude-opus-4-8$5.00$25.00200K
Claude Opus 4.7anthropic/claude-opus-4-7$5.00$25.00200K
Claude Opus 4.6anthropic/claude-opus-4-6$5.00$25.00200K
Claude Sonnet 4.6anthropic/claude-sonnet-4-6$3.00$15.00200K
Claude Sonnet 4.5anthropic/claude-sonnet-4-5-20250929$3.00$15.00200K
Claude Haiku 4.5anthropic/claude-haiku-4-5-20251001$1.00$5.00200K

Google

ModelModel IDInput ($/M)Output ($/M)Context
Gemini 3.1 Progoogle/gemini-3.1-pro$2.00$12.001M
Gemini 2.5 Flashgoogle/gemini-2.5-flash$0.30$2.501M
Gemini 2.5 Flash Litegoogle/gemini-2.5-flash-lite$0.10$0.401M
Gemini 3.1 Flash-Lite (Free)google/gemini-3.1-flash-lite$0.00$0.001M

xAI

ModelModel IDInput ($/M)Output ($/M)Context
Grok 4 Fast (Reasoning)xai/grok-4-fast-reasoning$0.20$0.502M
Grok Code Fastxai/grok-code-fast-1$0.20$1.50256K
Grok 3xai/grok-3$3.00$15.00131K
Grok 3 Minixai/grok-3-mini$0.30$0.50131K

DeepSeek

ModelModel IDInput ($/M)Output ($/M)Context
DeepSeek V3.2 Chatdeepseek/deepseek-chat$0.28$0.42128K
DeepSeek V3.2 Reasonerdeepseek/deepseek-reasoner$0.28$0.42128K
DeepSeek Coder V3deepseek/deepseek-coder$0.28$0.42128K

NVIDIA NIM

The free / oss / open profile routes here. Fifteen Nemotron models are free ($0/$0) — see Free tier for the per-tier mapping, or GET /v1/models for the full list. Two NVIDIA models are paid:

ModelModel IDInput ($/M)Output ($/M)Context
Llama 3.3 Nemotron Super 49B v1.5nvidia/nvidia/llama-3.3-nemotron-super-49b-v1.5$0.10$0.40128K
NVIDIA Nemotron Nano 9B v2nvidia/nvidia/nvidia-nemotron-nano-9b-v2$0.04$0.16128K

Note

These are provider list prices. The hosted gateway at api.solvela.ai has no platform fee, so the amount charged equals the provider cost. Always check the cost_breakdown in the 402 response for the exact amount you will pay. Self-hosted gateways may add a platform fee via SOLVELA_PLATFORM_FEE_PERCENT.

Query pricing programmatically

GET /pricing

Returns the full pricing table as JSON. No payment required.

curl https://api.solvela.ai/pricing

Cost estimation

The gateway estimates cost at request time using:

  1. Input tokens: estimated from the messages in the request (character count / 4, approximately)
  2. Output tokens: the completion-token ceiling — the tightest of max_tokens (if specified), the model's declared max_output_tokens, and a global 8,192-token cap. When max_tokens is omitted, the quote reserves for the full ceiling (up to 8,192 tokens), not a flat default
  3. Cost (on the hosted gateway, no fee): (input_tokens * input_rate + output_tokens * output_rate) / 1_000_000
  4. Cost (on self-hosted with fee enabled): (input_tokens * input_rate + output_tokens * output_rate) / 1_000_000 * 1.05

The actual cost charged via escrow claim is based on the real token usage returned by the provider (when available).

Example cost calculation

A request to gpt-4o with 500 input tokens and 200 output tokens, on the hosted gateway (fee suspended):

input_cost  = 500 / 1,000,000 * $2.50  = $0.00125
output_cost = 200 / 1,000,000 * $10.00 = $0.00200
provider_cost = $0.00325
platform_fee  = $0.00000 (suspended)
total         = $0.00325 USDC

Atomic amount: 3250 (multiply total by 1,000,000, round up).

With a self-hosted gateway at default 5% fee:

input_cost  = 500 / 1,000,000 * $2.50  = $0.00125
output_cost = 200 / 1,000,000 * $10.00 = $0.00200
provider_cost = $0.00325
platform_fee  = $0.00325 * 0.05 = $0.0001625
total         = $0.0034125 USDC

Atomic amount: 3413 (multiply total by 1,000,000, round up).