Pricing
Frontier and open models. One API. Priced below list.
Browse rates, estimate spend, compare providers, or use the CLI.
Pay with
Options at checkout depend on your country. Minimum top-up $5.
Published rates. A receipt on every call.
- Prepaid credit
- Top up from $5
- x402 per call
QuickSilver Machines — Agents ready out of the box. One unified bill for the machine and AI usage at normal model prices. No API keys or config files. $0.02 per running hour, billed per minute; paused is free; 200 running hours/month cap; free trial on the $0.05 signup credit, no card
81 models · Sorted by Popular
claude-fable-5-2coding, reasoning, and vision (5.2)
claude-sonnet-5-5newest Sonnet: faster, fewer tokens per task, strong coding
claude-haiku-5-5newest Haiku: small, fast, 1M context, built for subagents and high-volume work
claude-opus-5demanding reasoning, end-to-end coding, visual analysis, long-horizon agents
claude-sonnet-5newest Sonnet, stronger reasoning & coding, below list price
gpt-6.1-solGPT-6.1 Sol: near-Astra agentic coding and computer use at Sol pricing
everyday agentic coding, writing, and screenshot understanding
same model as Terra in pro reasoning mode — more reasoning tokens per request, same price
gpt-6-astralong-horizon agentic coding, deep research, document work, image input
gpt-6-solGPT-6 mid-flagship: strong reasoning, agentic coding, a rung below Astra
gpt-6-lunacheapest GPT-6 tier: high-volume chat, classification, lightweight agents
gpt-image-2.5-sunburstOpenAI GPT Image 2.5, precision tier, at medium quality
Coding, agents and mandatory reasoning with 1M context, tools and structured output
gemini-4-argonAlways-on reasoning, 1M context, text and images, tools and structured output
gemini-nano-banana-2.1Google's newest Flash-tier image generation and editing, 1K to 4K
deepseek-v4.1-flash-1017Long-context text and images, tools and structured output with optional reasoning
deepseek-v4.1-flashnewest DeepSeek Flash architecture, faster than V4 Flash
Meta 5 · StepFun 1 · xAI 1 · Alibaba / Qwen 10 · Z.ai 4 · Moonshot AI 4 · thinkingmachines 2 · MiniMax 1 · Xiaomi 3 · NVIDIA 1 · Tencent 2 · System One 2 · Black Forest Labs 3 · Stability AI 1 · ByteDance 2 · BRIA 2
muse-spark-openCoding agents, tools, structured output and image understanding
muse-wonderAgentic personal and professional work, mandatory reasoning, 1M context
step-5-previewReasoning, tools, structured output and image understanding
grok-4.7long-running agentic coding, knowledge work, self-verification, image input
qwen3.8-max-primeQwen top-tier flagship: deepest reasoning, autonomous coding
qwen3.8-flash-nextCheapest next-gen Qwen: 6B-active Qwen4 preview on a 1M-token window
qwen3.8-omni-flashQwen's first agentic omni model: image, audio + video understanding with tool use, at flash pricing
qwen3.8-27b-decenteredCommunity Macs for coding agents and long documents at 30% lower token prices.
kimi-k3.1-pro-highspeedLong-context coding and image-based workflows.
Lowest-priced Inkling for text, images and automatic tools.
Open-weight multimodal reasoning with 594B total and 17B active parameters.
Lowest-priced Tencent option: text, images and automatic tools
trio-spark-v1.3System One typed decisions on POST /v1/systemone: choice, score and noul with calibrated probabilities. Text input only on QSP, 1,024-token input limit (Jev: 32K). $0.042 per 1M input tokens; output free.
jev-1.13Typed decisions, not chat: routing, classification and scoring with calibrated probabilities via /v1/systemone
Prices are exact per-token rates, not rounded — some carry more decimals than others. Each struck-through figure is that row's reference list price. "At list" means we sell at the reference price; "mixed" means we are below it on one side and above on the other.
Every row checkable, every model verifiable →Building an agent? Fund a key programmatically with USDC — no account, no card.
x402 docs →Estimate a month
rates · catalog v87Plug in your monthly usage to see what it costs here, next to published list rates.
Model · 65 chat models
Typical prompt size
Models with a long-context rate switch to it above 200K prompt tokens. The breakdown shows the tier when it applies.
Monthly estimate for DeepSeek V4.1 Flash 1017
Thinking model — output token counts include the reasoning trace, which is typically 3-10× the visible reply.
QuickSilver Pro · per month
$6.37
Published list price
$6.37
Difference
$0.00
Same as the published list rate.
Monthly estimate for DeepSeek V4.1 Flash 1017: $6.37 per month
Basis QSP rates: catalog v87✓ catalog verified 2026-10-19
Same tokens at other providers
- Published reference$6.37 +0%
Same tokens on other models
Estimates use the selected model’s base input and output rates and your monthly token totals; adjust the sliders to your workload.
CLI qsp
Built for terminals and AI agents. --json output with stable exit codes — Claude Code, Cursor, Aider can call it without parsing HTML.
brew install machinefi/qspro/qspro
qsp init
qsp chat "Write a haiku about debugging." -m deepseek-v4.1-flash-1017How a dollar becomes a receipt
- Top up creditCard, wallet or USDC. $0.05 starter credit · no card.
- Call any modelTokens, images, audio seconds and characters, each at its published unit rate.
- Get usage.cost backEvery response reports its own cost, computed from those rates.
- Reconcile anytimeCheck against /pricing.json. Price changes are dated in the changelog.
Rates that aren’t one number
Long-context tiers, flat-rate speech and per-image pricing, stated in full. The same conditions appear inline in Models.
Long context · prompt > 200K tokens
Speech
OpenAI list prices: whisper-1 $0.006 per minute; tts-1 $15 per 1M characters.
Images · per image
+ 9 more image modelsChange the base URL. Keep your code.
The official OpenAI SDKs work as they are — point them at QuickSilver Pro.
# Two lines. That is the whole migration. from openai import OpenAI client = OpenAI( base_url="https://api.quicksilverpro.io/v1", api_key="your-api-key", )
FAQ
Common questions
QuickSilver Pro is an OpenAI-compatible inference API with 86 models, one endpoint and one API key. Browse models →
Yes. To turn reasoning off for direct chat, send reasoning.enabled=false in the request body; in Python use extra_body.
Claude Opus 5.5 costs 75% below Anthropic list. Rates vary by model and token type. DeepSeek V4.1 Flash 1017: $0.098 / $0.294. Kimi K3.1 Pro: $2.00 / $10.00. GLM 5.3: $0.60 / $2.20. Claude Fable 5.2: $2.00 / $10.00. GPT-6.2 Sol: $2.00 / $10.00. Grok 4.7: $2.00 / $6.00. Compare current catalog rates before choosing a model.
Yes. Change base_url to https://api.quicksilverpro.io/v1 in the official openai Python / Node / Swift SDKs. Streaming, tool calling, and usage.cost accounting all work out of the box. json_schema strict mode is model-dependent: the Claude models do not support it, so a schema there is advisory and the enforced path is a tool with strict: true.
New accounts get $0.05 free credit with no card, usable on 5 starter chat models in browser chat or through the API with your account key. API access also includes five speech models; top up from $5 to unlock the rest. Browser chat also has a separate allowance of 5 free messages. And your first credit purchase is matched 100%, up to $50 in bonus credits. The bonus is added to your balance the next time you top up (any amount from $5): pay $20, then top up $5 later, and $20 of bonus arrives with it. A larger first payment still receives the $50 maximum. One per customer and payment card; standard pay-as-you-go after that.
