Models
All prices below are at-cost — what the upstream provider charges us, we charge you. No markup, no spread. Last updated 2026-06-19.
Curated models
The 12 models below are the ones most customers reach for. They cover the bulk of real workloads (chat, code, vision, image generation).
Claude (Anthropic Messages)
| Model | Input / 1M tok | Output / 1M tok | Cache read |
|---|---|---|---|
claude-opus-4-8 | $15.00 | $75.00 | $1.50 |
claude-sonnet-4-6 | $3.00 | $15.00 | $0.30 |
claude-sonnet-4-5 | $3.00 | $15.00 | $0.30 |
claude-haiku-4-5 | $0.80 | $4.00 | $0.08 |
GPT (OpenAI Chat / Responses)
| Model | Input / 1M tok | Output / 1M tok | Cache read |
|---|---|---|---|
gpt-5.5 | $5.00 | $20.00 | $0.50 |
gpt-5.4 | $2.50 | $10.00 | $0.25 |
gpt-5.3-codex | $3.00 | $12.00 | — |
gpt-5.4-mini | $0.40 | $1.60 | $0.04 |
gpt-4o-mini | $0.15 | $0.60 | $0.075 |
Gemini (Google)
| Model | Input / 1M tok | Output / 1M tok | Cache read |
|---|---|---|---|
gemini-3.1-pro-preview | $3.50 | $10.50 | $0.875 |
gemini-3-flash-preview | $0.30 | $1.20 | $0.075 |
Image generation
| Model | Per image | Notes |
|---|---|---|
gpt-image-2 | $0.040 | 1024×1024, billed per generated image |
150+ more models available
The list above is curated for visibility — the full catalog is much larger. If you need a model that's not on the page (specific *-codex snapshot, embedding model, audio model, etc.), it almost certainly works. Email support@sparkapi.cc with the model ID and we'll confirm or add it.
Prompt caching
For long system prompts or repeated context blocks, enable prompt caching. Cached tokens are billed at roughly 10% of the input price — typically a 60–80% cost reduction on workloads with stable context.
How to enable it:
Claude (Anthropic Messages)
Add "cache_control": {"type": "ephemeral"} to any message or content block. Tokens above that breakpoint are cached for 5 minutes.
{
"model": "claude-sonnet-4-6",
"max_tokens": 256,
"system": [{
"type": "text",
"text": "<your long system prompt>",
"cache_control": {"type": "ephemeral"}
}],
"messages": [{"role": "user", "content": "summarise the system prompt"}]
}OpenAI
OpenAI auto-caches any prompt longer than 1024 tokens — no flag needed. The first request pays full price; subsequent requests within the cache window pay the cache-read rate from the table above.
Gemini
Pass "cachedContent": "<your-cache-name>" in the request body. See Google's docs for creating a cache first.