Full-spec Claude API
Stable, reliable, ready to use
FAQ
What request content and uses are prohibited?
You may not use the service to generate, request, or distribute the following: · Pornographic, sexually explicit, or other NSFW content · Graphic violence or gore · Hate speech, or content that promotes discrimination, harassment, or violence · Child sexual abuse material (CSAM), or any content that sexualizes, exploits, or endangers minors · Non-consensual deepfakes or other synthetic media intended to mislead or deceive · Content that infringes copyright, trademark, or other intellectual property · Political speech · Gambling or betting content · Scams or fraudulent content You also may not use the service or its outputs for model distillation or training — including training, fine-tuning, distilling, or otherwise developing or improving any AI model, and systematically collecting outputs to build training, distillation, or evaluation datasets. Distillation is a material breach: the account is immediately and permanently terminated, and remaining credits are forfeited without refund. Other violating requests may be blocked; serious or repeated violations can likewise result in account suspension or permanent termination without refund of remaining credits. See the Terms of Service for the full policy.
Is there a free trial or a minimum top-up?
The minimum top-up is just $10. No monthly fees, your balance never expires, no commitments. You are only charged for the tokens you actually use.
Is the service reliable? Is my data safe?
In stable operation since 2025 and available 24/7. Conversation content is used only to fulfil your requests and stays private. Optional TOTP two-factor authentication protects your account.
Do you modify my requests or inject a system prompt?
No. Your requests are forwarded verbatim — no prompt rewriting, no injected instructions, no model swapping, no post-processing; your own system prompt and tool definitions pass through unchanged. Forwarding may attach a small amount of standard metadata, which is served from the prompt cache automatically, so its cost is near zero in ongoing conversations.
A third-party authenticity checker shows one failed item — is the model degraded?
No. Requests reach Anthropic through an official channel and responses come back byte-for-byte, so you get the same full-spec model as official. A few checks (such as forced structured output) show unsupported because of a capability difference in that official channel itself — not a counterfeit or degraded model. Official fields like thinking signatures are present in every response and can be verified yourself.
