Unused credits are refundable within 30 days of purchase. Contact support with your account email and the unused balance goes back to your original payment method.

What are the rate limits?

Fair use, with defaults most workloads never hit. If you run sustained high-volume traffic and need reserved throughput, contact us.

What counts as cached input?

Parts of your prompt the servers have seen recently, such as a repeated system prompt. Cache hits are reported in usage.prompt_tokens_details.cached_tokens and billed at the lower rate automatically. Claude uses explicit cache_control blocks; support varies by model.

Are these the real models?

Receipts let you verify the model ID, token counts, and content hashes signed by Draftworks. They are not independent provider attestations.

What payment methods do you accept?

Cards, processed by Stripe. Purchases are one-time credit top-ups starting at $5. Nothing recurring is stored against your account.

Pricing

Cheaper on every token.

Credits are prepaid inference, spent at your tier's rates on any model in the catalog. There is no subscription and no commitment beyond the first $5.

Per 1M tokens	Provider list	Draftworks (−40%)
GPT-6 Astragpt-6-astra
Input	$10.00	$6.00
Cached input	$1.00	$0.60
Output	$50.00	$30.00
Cache writes (5 min)	$12.50	$7.50
Long-context input (over 272K)	$20.00	$12.00
Long-context cached input	$2.00	$1.20
Long-context output	$75.00	$45.00
Claude Fable 5.1claude-fable-5-1
Input	$10.00	$6.00
Cached input	$0.25	$0.15
Output	$50.00	$30.00
Cache writes (5 min)	$12.50	$7.50
Cache writes (1 hour)	$20.00	$12.00
Claude Opus 5claude-opus-5
Input	$5.00	$3.00
Cached input	$0.50	$0.30
Output	$25.00	$15.00
Cache writes (5 min)	$6.25	$3.75
Cache writes (1 hour)	$10.00	$6.00
Claude Sonnet 5claude-sonnet-5
Input	$2.00	$1.20
Cached input	$0.20	$0.12
Output	$10.00	$6.00
Cache writes (5 min)	$2.50	$1.50
Cache writes (1 hour)	$4.00	$2.40
Claude Fable 5claude-fable-5
Input	$10.00	$6.00
Cached input	$1.00	$0.60
Output	$50.00	$30.00
Cache writes (5 min)	$12.50	$7.50
Cache writes (1 hour)	$20.00	$12.00
Grok 4.6grok-4.6
Input	$2.00	$1.20
Cached input	$0.50	$0.30
Output	$6.00	$3.60
Long-context input (at least 200K)	$4.00	$2.40
Long-context cached input	$1.00	$0.60
Long-context output	$12.00	$7.20
Kimi K3kimi-k3
Input	$3.00	$1.80
Cached input	$0.30	$0.18
Output	$15.00	$9.00
GLM 5.3glm-5.3
Input	$1.40	$0.84
Cached input	$0.26	$0.156
Output	$4.40	$2.64
GPT-5.6 Solgpt-5.6-sol
Input	$4.00	$2.40
Cached input	$0.40	$0.24
Output	$20.00	$12.00
Cache writes (5 min)	$5.00	$3.00
Long-context input (over 272K)	$8.00	$4.80
Long-context cached input	$0.80	$0.48
Long-context output	$30.00	$18.00
GPT-5.6 Terragpt-5.6-terra
Input	$2.00	$1.20
Cached input	$0.20	$0.12
Output	$12.00	$7.20
Cache writes (5 min)	$2.50	$1.50
Long-context input (over 272K)	$4.00	$2.40
Long-context cached input	$0.40	$0.24
Long-context output	$18.00	$10.80
GPT-5.6 Lunagpt-5.6-luna
Input	$0.20	$0.12
Cached input	$0.02	$0.012
Output	$1.20	$0.72
Cache writes (5 min)	$0.25	$0.15
Long-context input (over 272K)	$0.40	$0.24
Long-context cached input	$0.04	$0.024
Long-context output	$1.80	$1.08
GPT-5.4 Progpt-5.4-pro
Input	$30.00	$18.00
Cached input	—	—
Output	$180	$108
Grok 4.3grok-4.3
Input	$1.25	$0.75
Cached input	$0.20	$0.12
Output	$2.50	$1.50
Kimi K2.7 Codekimi-k2.7-code
Input	$0.95	$0.57
Cached input	$0.19	$0.114
Output	$4.00	$2.40
Kimi K2.7 Code Fastkimi-k2.7-code-fast
Input	$1.05	$0.63
Cached input	$0.18	$0.108
Output	$4.40	$2.64
GPT Image 2gpt-image-2 · Scale tier only
Text input	$5.00	$3.00
Cached text input	$1.25	$0.75
Image input	$8.00	$4.80
Cached image input	$2.00	$1.20
Image output	$30.00	$18.00
GPT Realtime 2.1gpt-realtime-2.1 · Scale tier only
Text input	$4.00	$2.40
Cached text input	$0.40	$0.24
Text output	$24.00	$14.40
Audio input	$32.00	$19.20
Cached audio input	$0.40	$0.24
Audio output	$64.00	$38.40
Image input	$5.00	$3.00

The discount applies to every model, no exceptions. Context windows and versions are on the models page. GPT-5.4 Pro has no cached input rate.

Enterprise deals

Have a larger workload or specific requirements? Talk to us about a custom proposal.

Explore enterprise

Scale tier

Unlocks image generation and realtime voice once lifetime purchases reach $5,000, with the same 40% discount. Access is permanent; no minimum ongoing spend is required.

Credits

One credit is one dollar of inference at your tier's rates. The minimum purchase is $5, and unused credits are refundable within 30 days.

Estimate your monthly cost.

Drag the sliders to your expected volume.

Model

Input tokens / month50M

Output tokens / month10M

Provider list

$1,000.00

Draftworks

$600.00

−$400.00/mo

Assumes uncached input at standard context rates. Long-context requests and cache writes cost more. Cached input for gpt-6-astra is billed at $0.60 per 1M on both tiers.

Common questions.

Can I get a refund?: Unused credits are refundable within 30 days of purchase. Contact support with your account email and the unused balance goes back to your original payment method.
What are the rate limits?: Fair use, with defaults most workloads never hit. If you run sustained high-volume traffic and need reserved throughput, contact us.
What counts as cached input?: Parts of your prompt the servers have seen recently, such as a repeated system prompt. Cache hits are reported in usage.prompt_tokens_details.cached_tokens and billed at the lower rate automatically. Claude uses explicit cache_control blocks; support varies by model.
Are these the real models?: Receipts let you verify the model ID, token counts, and content hashes signed by Draftworks. They are not independent provider attestations. Check a receipt on the verify page.
What payment methods do you accept?: Cards, processed by Stripe. Purchases are one-time credit top-ups starting at $5. Nothing recurring is stored against your account.

Start with $5.Priced like infrastructure.

Create an account Read the quickstart

Unused credits are refundable for 30 days.