Draftworks

Models

The GPT-5.6 family, GPT-5.4 Pro, Grok 4.3, Kimi K2.7 Code, image generation, and realtime voice, every one of them 20-25% under provider list.

Draftworks serves models from OpenAI, xAI, and Moonshot AI through one API and one billing pipeline. Every model is priced 20% under provider list on the Base tier and 25% under on the Scale tier, and every response ships a signed receipt.

GPT-5.6 is live

OpenAI released the GPT-5.6 family on July 9, 2026, and Sol, Terra, and Luna were serving on Draftworks the same day. They replace GPT-5.5, which has been retired from the catalog; gpt-5.6-sol is its direct successor at identical list pricing.

List models
curl https://api.draftworks.dev/v1/models

Catalog

Model idVendorContextEndpointTier
gpt-5.6-solOpenAI1MChat Completions, ResponsesAll
gpt-5.6-terraOpenAI1MChat Completions, ResponsesAll
gpt-5.6-lunaOpenAI1MChat Completions, ResponsesAll
gpt-5.4-proOpenAI1MResponses onlyAll
grok-4.3xAI200KChat CompletionsAll
kimi-k2.7-codeMoonshot AI256KChat CompletionsAll
kimi-k2.7-code-fastMoonshot AI256KChat CompletionsAll
gpt-image-2OpenAIImagesScale
gpt-realtime-2.1OpenAIRealtimeScale

Pinned snapshots are also accepted as model ids where the vendor publishes them, for example gpt-5.6-sol-2026-07-09. GET /v1/models always returns the current list, including the pinned alias in root.

Rates

Per 1M tokens. List is the provider's published price; you pay the Base or Scale column.

Language models

ModelRateListBase (−20%)Scale (−25%)
gpt-5.6-solInput$5.00$4.00$3.75
Cached input$0.50$0.40$0.375
Output$30.00$24.00$22.50
gpt-5.6-terraInput$2.50$2.00$1.875
Cached input$0.25$0.20$0.187
Output$15.00$12.00$11.25
gpt-5.6-lunaInput$1.00$0.80$0.75
Cached input$0.10$0.08$0.075
Output$6.00$4.80$4.50
gpt-5.4-proInput$30.00$24.00$22.50
Output$180$144$135
grok-4.3Input$1.25$1.00$0.937
Cached input$0.20$0.16$0.15
Output$2.50$2.00$1.875
kimi-k2.7-codeInput$0.95$0.76$0.712
Cached input$0.19$0.152$0.142
Output$4.00$3.20$3.00
kimi-k2.7-code-fastInput$1.05$0.84$0.787
Cached input$0.18$0.144$0.135
Output$4.40$3.52$3.30

GPT-5.4 Pro does not offer cached input.

Image and voice

ModelRateListBase (−20%)Scale (−25%)
gpt-image-2Text input$5.00$4.00$3.75
Image input$8.00$6.40$6.00
Image output$30.00$24.00$22.50
gpt-realtime-2.1Text input$4.00$3.20$3.00
Text output$24.00$19.20$18.00
Audio input$32.00$25.60$24.00
Audio output$64.00$51.20$48.00
Image input$5.00$4.00$3.75

Cached input for these models follows the same 20/25% discount off the provider's cached rate. Realtime audio tokens are duration-encoded by the upstream.

Choosing a GPT-5.6 model

The three variants share the 1M context window and differ in size and price. gpt-5.6-sol is the flagship for reasoning-heavy and agentic work. gpt-5.6-terra holds most of that quality at half the price and suits production workloads with volume. gpt-5.6-luna is the small, fast option for classification, extraction, and other high-throughput tasks at $0.80 per 1M input on the Base tier.

If you were on gpt-5.5, move to gpt-5.6-sol: same list price, same endpoints, newer model. Requests that still name gpt-5.5 return 404 model_not_found with the current model list.

Kimi fast mode

kimi-k2.7-code and kimi-k2.7-code-fast are the same K2.7 Code weights on different serving stacks. The fast deployment is tuned for interactive latency and costs about 10% more; the standard deployment is the cheaper default for batch and background work. Responses from both report the public model id you requested.

Scale tier models

gpt-image-2 and gpt-realtime-2.1 require the Scale tier, which unlocks automatically and permanently once lifetime purchases reach $5,000. Calls from Base tier accounts return:

{
  "error": {
    "message": "The model `gpt-image-2` is available on the Scale tier. The Scale tier unlocks automatically once lifetime purchases reach $5,000. See /pricing.",
    "type": "invalid_request_error",
    "code": "model_requires_scale_tier"
  }
}

Choosing an endpoint

  • Text models take POST /v1/chat/completions, except gpt-5.4-pro which is Responses API only; the GPT-5.6 family supports both.
  • Requesting a model on an endpoint it does not support returns 404 model_not_supported with the endpoint it lives on.

On this page