ProselonWhich model should I pick from OpenRouter?
Any model works with Proselon's API Key engine so long as it can call tools, because the co-writer reaches your chapters by calling them. Every model in the table below could on the day we checked. Click a column heading to sort by it.
Which models are best for Proselon?
| Model | In $/M | Out $/M | Context | Unfiltered |
|---|---|---|---|---|
GLM 5.3 Flashz-ai/glm-5.3-flashOpen weights. Third most-used. | $0.09 | $0.30 | 1.31M | Yes |
DeepSeek V4.1 Flashdeepseek/deepseek-v4.1-flashOpen weights. Fifth most-used. | $0.15 | $0.60 | 1.05M | Yes |
GPT-5.6 Lunaopenai/gpt-5.6-lunaThe most-used model. | $0.20 | $1.20 | 1.05M | No |
Gemini 3.8 Flashgoogle/gemini-3.8-flashGoogle's Flash model. | $0.75 | $3.75 | 1.05M | Yes |
Hy4 previewtencent/hy4-previewOpen weights. Second most-used. | $0.83 | $2.50 | 1.05M | Yes |
Claude Haiku 4.5anthropic/claude-haiku-4.5Anthropic's cheapest. | $1.00 | $5.00 | 200K | No |
Grok 4.3x-ai/grok-4.3Longer context than 4.6, lower price. | $1.25 | $2.50 | 1M | Yes |
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-previewGoogle's Pro model. | $2.00 | $12.00 | 1.05M | Yes |
Grok 4.6x-ai/grok-4.6xAI's flagship. | $2.00 | $6.00 | 500K | Yes |
Kimi K3moonshotai/kimi-k3Open weights. | $3.00 | $15.00 | 1.05M | Yes |
Claude Fable 5.1anthropic/claude-fable-5.1Anthropic's flagship. | $10.00 | $50.00 | 1M | No |
GPT-6 Astraopenai/gpt-6-astraOpenAI's flagship. | $10.00 | $50.00 | 1.05M | No |
Every figure came from OpenRouter's public model list on 17 September 2026, and every one of them moves, so read the live page before you commit. Prices are per million tokens, read and written. Where a line under a model ranks it by use, the ranking is OpenRouter's own for the week to 16 September 2026.
What Unfiltered means
The Unfiltered column is OpenRouter's own flag. No means OpenRouter screens requests to that model before they arrive, at the vendor's requirement, and Yes means it passes them through untouched. An unfiltered model is the place to begin if you write explicit or violent fiction (what forums call an uncensored or NSFW model), though the column records only OpenRouter's screening, and a model can still decline a request by its own training. Proselon places no restriction of its own on what you write, and whatever limit you meet belongs to the provider.
What other writers run
In the week to 16 September 2026 the model that processed the most tokens on OpenRouter was GPT-5.6 Luna, and behind it came Tencent's Hy4 preview, GLM 5.3 Flash and two releases of DeepSeek V4 Flash. The view nearest to novel writing, which OpenRouter calls Roleplay & Fiction and puts at 3.8 per cent of all spend across those seven days, was led by Claude Opus 4.6, GLM 5.2 and DeepSeek V4 Flash, with four further Claude releases in its top ten. OpenRouter states on that page that these figures measure adoption rather than quality, and we repeat it.
A model that is not on this list
Open the model's page at openrouter.ai and read the list of parameters it supports. tools in that list means the model can run the co-writer, and its absence means it cannot, whatever its other merits. What you spend goes to OpenRouter at its own rates. Proselon is $6 a month and runs no second meter, so we add no markup and take nothing from your key.
If you would rather not run a key at all
Proselon's other engines sign you in with an account you already hold, whether Claude, ChatGPT, Grok or Gemini, and the writing then runs on the subscription you already pay for, with nothing metered and no model to name. Setting up your Writing Engine covers those. If you would rather have no bill at all, a model on your own computer is free and offline, and which local model works best with Proselon is the article for that.