Authors can use Sudowrite without ever changing the default settings, but it’s still helpful to know that Sudowrite does offer a variety of Prose Modes and AI models to suit different purposes.

Prose Modes are AI model suites specific to Sudowrite, designed specifically for fiction writing. These modes let you experience different output styles at varying costs, each offering unique balances of creativity, control, content filtering, and credit usage. We’ll dive in below on the Prose Modes and AI models available on Sudowrite.

Muse is Sudowrite's flagship proprietary model, purpose-built just for fiction. It was trained by writers for writers, unlike general AI models that try to "do everything." This focus allows Muse to achieve a higher level of story quality and nuance.

Prose Quality & Style: Muse delivers vivid, original prose with a strong creative voice, free from common AI clichés. Excels at rich descriptions, emotional beats, and maintaining consistent tone throughout your story.

Instruction Following: Prioritizes creative interpretation over strict obedience. Continues your story in imaginative ways while still incorporating Story Bible context and outline beats effectively for narrative continuity.

Content Filter Level: Virtually none. Handles any content from dark and gritty to light and humorous, including graphic or mature themes that other AIs might refuse. Use discretion, but know Muse won't shy away from adult content.

Credit Usage: Higher credit cost than Basic mode (comparable to former "Premium" setting). Many authors find the quality worth the expense, though consider switching to cheaper models for brainstorming or extremely long outputs.

Best Use Cases: Ideal for high-quality fiction prose, especially creative, character-driven writing. Excels in impactful scenes, character interactions, and climactic moments. Perfect for unfiltered content including mature themes. Outputs often require less editing than other models.

Weaknesses: Not optimized for non-fiction or highly structured tasks. Definitely voice-y, and perhaps not to everyone’s taste. May occasionally produce unexpectedly edgy content. Available exclusively within Sudowrite's ecosystem.

Excellent utilizes Anthropic's Claude 3.7 Sonnet model as its primary engine. This mode is named "Excellent" because it aims for an excellent balance between creative prose and obedience to your instructions.

Prose Quality: Produces well-structured, descriptive writing with a flair for action and character movement. While not as stylistically unique as Muse, delivers engaging narrative that typically needs minimal polishing.

Instruction Following: Excellent's standout feature. Claude follows complex instructions precisely while maintaining context. Adheres closely to outlines and beats while still expanding them creatively. Reliable for both storytelling and non-prose tasks like summarizing or brainstorming.

Content Filter Level: Moderate. More permissive than OpenAI models regarding violence and mild adult content, but has boundaries. May imply rather than explicitly describe adult situations or graphic content. Handles typical novel scenes well but might gloss over extremely explicit content.

Credit Usage: Mid-tier cost—more expensive than Basic but cheaper than Muse. Efficiently provides substantial output without rapidly depleting credits, making it ideal for budget-conscious writers who still want quality.

Best Use Cases: Perfect for general drafting and rewriting with balanced creativity and reliability. Excels with dialogue-heavy scenes, structured narratives, and when following outlines closely. Ideal for co-writing projects where story beats and world-building details must be respected. Works well for instruction-heavy tasks in plugins.

Weaknesses: Prose can feel less distinctive than Muse, sometimes "safer" or generic with minimal guidance. May faithfully reproduce errors in your outline or beats. Might not deliver the intensity needed for certain scenes in your first attempt, due to its caution with mature content.

Basic uses the lightweight and economical GPT-4o Mini as its backbone. This mode is all about speed and efficiency, giving you decent writing while stretching your credit budget. Don't let the name "Basic" mislead you – it's quite powerful, just tuned for cost-effectiveness.

Prose Quality: Delivers coherent, straightforward content that serves as solid first-draft material. Provides clear descriptions without poetic flourishes (e.g., "small and dimly lit room" vs. "bathed in faint flickering light"). Still remarkably competent for its size—better than older GPT-3.5 models with fewer errors, if less original.

Instruction Following: Highly obedient and literal, following beats and instructions closely. Stays on track without introducing unexpected plot elements. Limited "story sense" compared to larger models—won't infer beyond what's explicitly prompted, which can lead to generic output if guidance is minimal.

Content Filter Level: Strict, inheriting OpenAI's content moderation. Avoids explicit sexual content and extreme violence, often skipping or neutralizing such elements. Adequate for mainstream fiction but may refuse or severely water down mature content in erotica or intense horror.

Credit Usage: Exceptionally credit-efficient—Basic's standout feature. Generates substantial text at minimal cost, making it ideal for limited budgets or massive projects. Perfect for exploratory writing, generating multiple scene versions, or any high-volume text needs.

Best Use Cases: Excels at speed-writing, brainstorming, quick drafting, and plot development. Ideal for connective scenes, experimentation, and utility tasks (summarizing, listing ideas). A dependable workhorse for getting words on the page efficiently.

Weaknesses: Lacks sophisticated elements like metaphors or emotional subtext. Complex or character-rich scenes may feel flat. May fall back on cliches or tropes in cases. Best used for basics with plans to enhance later.

Sudowrite's Advanced selector offers AI models beyond the main Prose Modes, including Experimental options and Retired Prose Modes. These provide granular control for specific chapters or specialized plugin tasks. Here's a concise overview of available advanced models:

Sudowrite provides access to Anthropic's Claude 3 family in three flavors. Claude models are known for knowledgeable, thoughtful writing and a large context window, making them great for novelists who feed a lot of story context.

Claude 3.0 Opus: Once the most powerful Claude model (powers the Retired "Best Prose v2"), Opus is still preferred by many. Excels at highly detailed writing with seamless incorporation of Story Bible information and prior chapters. Perfect for complex scenes with callbacks to earlier foreshadowing or intricate settings. Follows sophisticated instructions precisely, producing richly detailed, cohesive output. Drawbacks include slower processing, very high credit usage, and occasional verbosity requiring trimming. Ideal for final polishing or ensuring continuity in complex narratives.

Claude 3.0, 3.5 and 3.7 Sonnet: Balances speed and quality. Produces high-quality writing with faster turnaround than Opus. Especially strong for action scenes and general chapter drafting with lively, focused descriptions that advance the plot efficiently. More credit-friendly than Opus while maintaining strong output quality requiring minimal editing. Watch for occasional formality in dialogue.

Claude 3.0, and 3.5 Haiku: The fastest Claude variant, offering a middle ground between speed and detail. Perfect for everyday writing tasks, transitional scenes, and rapid iteration. Follows prompts reliably with coherent output, though style is more straightforward. Benefits include speed and lower credit cost, making it excellent for experimentation. Less nuanced with complex or emotional content.

The Claude models are widely preferred, and work well for standard genre fiction with a thoughtful, mildly stylistic voice. While Claude models typically feature safety filters that avoid certain themes—sexual content, graphic torture, etc.—they are more permissive in the context of fiction writing on Sudowrite. Note that despite our bucketing here, the higher numbers assigned to these models by Anthropic indicate more recent iterations. That means Claude 3.7 Sonnet is the newest (and therefore most powerful) model available from Anthropic.

Sudowrite integrates OpenAI's GPT-4 Omni models, which are the evolution of GPT-4 with multimodal capabilities—though in the context of Sudowrite, they are text-only. GPT-4o models are extremely capable in terms of language understanding and generation. They are especially known for factual accuracy, logical coherence, and adherence to instructions, which can be very useful for certain writing scenarios.

GPT-4o (Full): One of the most versatile AI models available, GPT-4o excels at creating vivid, sensory-rich descriptions and detailed settings. It is particularly good at structured tasks, closely following outlines and maintaining consistency throughout long passages. It's also strong for analytical or complex narrative tasks. The downside includes relatively high credit costs and firm content restrictions—though these are somewhat relaxed within Sudowrite. GPT-4o tends to be cautious with mature content, and authors may need to explicitly prompt deeper emotional nuance. Ideal for scenes requiring precision, final consistency checks, or polished literary prose.

GPT-4o Mini: A streamlined, cost-effective variant that powers Sudowrite's Basic mode. GPT-4o Mini delivers impressive quality at reduced cost, producing reliable and clear descriptions without the lush details of its larger counterpart. It maintains strong obedience to instructions and handles large context windows well, ensuring narrative consistency. Particularly useful for general writing tasks, drafting, and plugins, offering a dependable balance of quality and economy.

GPT-4.1 (Full): GPT-4.1 stands out for its superior ability to follow intricate instructions right from the first draft, making it especially beneficial for authors who need precise adherence to complex plot details. Its remarkable one-million-token context window greatly improves consistency, allowing detailed story continuity across extended narratives—far beyond typical novel lengths. The model's robust story understanding results in fewer required edits and tighter adherence to your vision. GPT-4.1 offers excellent value, often outperforming GPT-4 Turbo (Balanced) in terms of cost-effectiveness and quality, making it suitable for detailed, long-form writing tasks and critical scenes demanding precision.

GPT-4.1 Mini: The "brainy" younger sibling of GPT-4.1, this model provides substantial quality at roughly twice the cost of GPT-4o Mini but with notably improved instruction-following and context retention. It's ideal for scenes or chapters requiring a careful balance of creativity and fidelity to detailed outlines or story elements. GPT-4.1 Mini delivers reliable, consistent outputs at an appealing price point for authors who want quality without the full cost of the premium models.

GPT-4.1 Nano: GPT-4.1 Nano is remarkably cost-effective—approximately 30% cheaper than GPT-4o Mini—yet still benefits from the GPT-4.1 suite's enhanced instruction-following and memory capabilities. It's perfect for high-volume drafting, brainstorming, and preliminary scene-building, where authors can generate large amounts of text affordably before refining it with a higher-tier model. Though output is simpler and less detailed than its bigger siblings, its economy and improved narrative coherence make it an exceptional value for authors on tighter budgets.

Both GPT-4o and GPT-4.1 suites generally produce precise, grammatically polished text, though the tone may sometimes feel formal or academic. They respond exceptionally well to stylistic guidance, quickly adapting when provided with clear examples of the desired voice or tone.

DeepSeek is a family of cutting-edge open-source models that Sudowrite has added, especially appealing to those looking for cost-effective alternatives to the big-name APIs. DeepSeek models are trained with novel techniques emphasizing reasoning and economy. They are relatively new (introduced in 2025) and have quickly become popular for certain use cases.

DeepSeek V3: Optimized for maximum value—offering performance comparable to larger models at a significantly lower cost. Provides competent, "serviceable" prose with solid grammar and basic creativity. Functions as a blank stylistic canvas you can shape. Even cheaper than GPT-4o Mini, making it among Sudowrite's most credit-efficient option for long-form content. Ideal for quick brainstorming, rough drafts of entire novels on limited budgets, or generating large volumes of content for later enhancement. Output may occasionally contain awkward phrasing but nothing Quick Edit can't fix.

DeepSeek R1: Specializes in reasoning power with an internal "thinking" phase before writing. Excels with complex narrative problems, convoluted plots, mysteries with numerous clues, or maintaining character consistency. Produces coherent, logically consistent chapters where characters stay in character and plot threads

Sudowrite also includes a few other models that cater to specific needs or user preferences. These might not be the top choices for most, but they can be useful in specific scenarios or for experimentation.

Mythomax 13B: A 13-billion-parameter open-source model fine-tuned for storytelling. Valued for speed and decent dialogue capabilities. Perfect for quickly generating scene drafts to overcome writer's block. Output is adequate but not deeply polished—expect to rewrite portions. Excellent for casual brainstorming and character dialogue exploration. Sometimes provides creative, unconventional ideas mainstream models might avoid.

Mistral 7B (Mixtral): A very small, extremely fast model described as "well-rounded" for new users. Limited to basic tasks like sentence completion or simple rewrites. Produces flat, sometimes repetitive prose over longer passages—not recommended for actual chapter writing. Useful for quick sentence rephrasing or testing continuations without consuming significant credits. Its smaller size means less bias but also less knowledge. Consider it an experimental option, as serious fiction work typically requires larger models.

Goliath 120B: A legacy 120-billion parameter model formerly used as Sudowrite's "Unfiltered" Prose Mode before Muse. Creates lengthy, coherent text with a permissive approach to content. Known for natural, conversational tone and potential for humor. Computationally heavy (slower) and less finely tuned than newer models. Sometimes produces shorter outputs requiring extra prompting to continue. With Muse now available, Goliath serves mainly as an alternative voice—its straightforward pulp-fiction style might suit certain scenes, but Muse or DeepSeek V3 typically outperform it on quality and cost.

(Others): Sudowrite's model lineup continues evolving. You may see additions like Google's Gemini Pro or other open-source options in the future. Check Sudowrite's changelog for new additions, which typically include usage notes. Models labeled "Experimental" have been implemented and tested, but not exhaustively and results may still vary in different contexts.

Choosing the right AI model in Sudowrite significantly impacts your writing experience. For most fiction writers, start with the main Prose Modes:

As you grow comfortable with the platform, explore Advanced models for specific needs—Claude 3.7 Sonnet for intricate chapters requiring detail and continuity, DeepSeek Instant for breaking through creative blocks, or Mythomax for quick scene drafts.

Remember that these models serve as collaborators, not replacements. Your voice, editing skill, and creative vision remain irreplaceable. The key is matching the right model to each specific task:

By understanding each model's strengths and limitations, you can transform Sudowrite into a personalized writing partner that adapts to different phases of your creative process. Happy writing!