Open studio

AI models

Configuration effective for this version: 2026-10-26. Skald Studio uses OpenAI models. Standard and Advanced describe your AI choice; your Free, Writer, Author or Studio subscription controls which features and choices are available.

Standard

Default Standard model by feature
FeatureRequested model ID
Brainstormgpt-6-luna
Foundation researchgpt-6-luna
Outlinegpt-6-luna
Character rostergpt-6-luna
Character chatgpt-6-luna
Agent Skald chatgpt-6-luna
Chapter plangpt-6-luna
Chapter draftgpt-6-luna
Selected passage revisiongpt-6-luna
Locationsgpt-6-luna
Relationship webgpt-6-luna
Relationship arcgpt-6-luna
Relationship scangpt-6-luna
Field suggestionsgpt-6-luna

Standard uses gpt-6-luna with reasoning effort none. It is available within each plan’s existing feature access and credit allowance. Retired text choices are unavailable; they are not automatic fallbacks.

Advanced

Existing opt-in Advanced actions use gpt-6.1-sol with low reasoning effort. They consume Premium credits in addition to the displayed Standard credit cost. Feature eligibility and monthly limits still apply; explicit Advanced chat selection requires Studio. A plan upgrade never automatically selects Advanced.

Evaluation and other tools

Selected writing comparisons and completeness checks support these choices. Results vary with the task and context; these checks do not establish error-free output. Research, scans and long drafts may take longer than other tools.

Character portraits request gpt-image-2.5-flare-2026-09-08, separately from text choices. Manuscript evaluation is currently unavailable; its internal model mapping does not enable the feature.

Portrait configuration effective 2026-10-26: requests pin this snapshot, with no automatic model fallback. The provider may omit the returned model; in that case the actual served model remains unknown. We retain the pinned request and response evidence without claiming provider verification. Low and high quality keep their existing displayed credit charges.

Controlled-beta portrait evaluation covers a low-quality Watercolor sample. High-quality Cinematic Realism and other styles have not received equivalent quality and cost evaluation. Styles and plan access remain available; review each result for visual consistency with your character. Observed provider costs are estimates, not guaranteed maximum charges.

Fallbacks and attribution

Omitting a model selects that feature’s Standard default. An unknown internal feature resolves to gpt-6-luna, but cannot run without a defined spending policy. Unsupported explicit model requests are rejected by the text gateway. Provider failures do not silently switch to another model. The outline tool may make one bounded structural repair call with the same model.

A Standard fiction chapter outside the requested length may receive one correction using the same model, included in its original credit quote. Incomplete results are rejected. This does not authorize repeated retries or change accepted canon.

These are provider aliases and may resolve to dated versions. Internal provider usage records retain requested and returned model IDs when supplied, token usage and cost uncertainty. Incomplete or interrupted requests can still incur provider cost; this is tracked separately from customer credit refunds. The displayed quote determines your credit charge.

Text generation currently receives a complete provider response before delivering it in chunks to the editor. Model output can miss length, invent details or make factual errors. Review facts, continuity and edits before accepting them. Models do not automatically save accepted canon.

How credits work · Privacy