This update applies to AI Gateway, not to v0 itself. Vercel's v0 pages do not mention Fable 5.2, so the announcement does not mean v0 users should expect to find the model in the editor. See the v0 section below for details.
For anyone building sites with AI tools, the practical details are simple: the price matches the provider's price, with no markup or platform fee, and using the model comes with two conditions. Prompts are stored for 30 days, and safety checks may decline some requests.
What is new
- Claude Fable 5.2 is available on AI Gateway from November 9, 2026, under the model name anthropic/claude-fable-5.2.
- You can call it through the AI SDK, the Chat Completions, Responses and Anthropic Messages APIs, or a coding agent connected to AI Gateway.
- Reasoning effort runs from low to max, and high is the API default.
- Cybersecurity and biology safety classifiers are switched on, and you can set fallback models for requests a classifier declines.
- Using the model requires 30-day retention of prompts and completions.
Taken from the official update: vercel.com [1]
What this means for v0 users
The v0 pricing page says Plus, Business and Enterprise plans include access to all models, while the v0 changelog says the model picker can browse AI Gateway models. Fable 5.2 is available on AI Gateway, but v0's own pages do not mention it, so they do not confirm that it can be selected in v0.
Fact checked on 10 Nov 2026 against v0.app [4], v0.app [5]
The model that v0's changelog does mention in the picker is Claude Opus 5. The changelog also says v0 Max uses Claude Opus 4.8 and Max Fast uses Opus 4.8 Fast. For people working in v0, those are the models described on its pages today. This announcement applies to people calling models from their own app or coding agent.
Fact checked on 10 Nov 2026 against v0.app [4]
What Fable 5.2 is meant for
Vercel describes Fable 5.2 as a model for open-ended technical work, with improvements over earlier Fable models in coding, reasoning and vision. It can also handle long, unclear tasks unattended. For a site builder, that could mean planning a change across many files or interpreting a layout screenshot.
Fact checked on 10 Nov 2026 against vercel.com [1]
The effort setting lets you balance depth and speed. It ranges from low to max, with high as the API default if you make no selection. Use low for quick, simple requests and reserve the higher levels for tasks where a wrong answer would cost time.
Fact checked on 10 Nov 2026 against vercel.com [1]
How to call it and what it costs
The model name to use is anthropic/claude-fable-5.2. Vercel says it is available through the AI SDK (Vercel's code toolkit for AI features), the Chat Completions, Responses and Anthropic Messages APIs (standard methods for a program to send requests to an AI model), or a coding agent connected to AI Gateway. To use Claude Code, Codex, Cursor or another supported agent, Vercel says to install the latest Vercel CLI and run npx vercel ai-gateway setup. The Claude Code documentation also gives this command: npx vercel@latest ai-gateway setup --agent claude-code.
Fact checked on 10 Nov 2026 against vercel.com [1]
AI Gateway's model catalog lists Fable 5.2 at $8 per million input tokens and $40 per million output tokens. Tokens are small pieces of text that a model reads and writes. Vercel says AI Gateway passes through the provider's pricing without a markup or platform fee for inference, including requests made with your own key, meaning a key you already have from the model maker.
Fact checked on 10 Nov 2026 against vercel.com [3], vercel.com [1]
For comparison, AI Gateway lists the earlier Claude Fable 5.1 at $10 per million input tokens and $50 per million output tokens. Fable 5.2 costs $2 less per million input tokens and $10 less per million output tokens, a 20 percent reduction for both. A short prompt costs only a small fraction of a dollar either way, but long agent runs can add up, so estimate the cost before letting an agent loop.
Fact checked on 10 Nov 2026 against vercel.com [3], vercel.com [6]
Every Vercel plan includes AI Gateway. The pricing documentation says the free tier supports only eligible models, while the paid tier provides access to every available model. So a free account might not be able to call Fable 5.2, and you should check the model list for your account before making plans around it.
Fact checked on 10 Nov 2026 against vercel.com [2]
Data retention and safety checks to plan for
Using Fable 5.2 through AI Gateway requires data retention. Prompts and completions are stored for 30 days and are not used to train Claude. If Zero Data Retention is enabled for your team, AI Gateway returns an error rather than sending the request. Anthropic offers Zero Data Retention to some customers under separate agreements, but that does not change how AI Gateway handles this model.
Fact checked on 10 Nov 2026 against vercel.com [1]
Check this first if you build sites for clients who handle private customer data. Do not send client records, passwords or personal details through this model unless the client has agreed to the retention.
Cybersecurity and biology safety classifiers are also enabled for the model. If the cybersecurity classifier declines a request, you can keep your app working by adding a models array under providerOptions.gateway, the setting for listing backup models. AI Gateway then tries the models in order. The announcement's example uses Opus 5.5 as the first fallback and Sonnet 5.5 next, and says this option works the same way with every AI Gateway API format.
Fact checked on 10 Nov 2026 against vercel.com [1]
What the official notes do not say
- The announcement and the linked pricing page do not say whether Claude Fable 5.2 can be used with the free tier credits.
- The official pages do not give a fixed rate limit in numbers for Claude Fable 5.2.
- The announcement says Fable 5.2 improves on earlier Fable models but does not say it replaces a particular model or version.
- The v0 pricing and changelog pages do not name Claude Fable 5.2, so they do not say whether it is offered in v0 or on which v0 plans.
How it helps: a freelancer who maintains sites for local shops and works with a coding agent
The task. Plan a change affecting many files on a client's site, such as shifting the shop's menu pages to a new layout
- Check the client's data rules first, because a team with Zero Data Retention turned on gets an error for this model and everything you send is kept for 30 days.
- Install the latest Vercel CLI and run npx vercel ai-gateway setup, then choose your coding agent.
- Set the model name to anthropic/claude-fable-5.2 in the agent.
- Choose the effort per task: low for a quick question about one page, and high, the default, for the multi-file plan.
- Estimate the cost from the listed $8 and $40 per million tokens before you let the agent run a long job, since AI Gateway adds no markup on top of that price.
The result. A model that Vercel describes as suited to long, open-ended work gives you a change plan through the same AI Gateway connection as your other models. You know the cost and the 30-day retention before you begin.
Fact checked on 10 Nov 2026 against vercel.com [1], vercel.com [3]
Other ways to use it
Free access includes only eligible models, and the documentation does not say whether Fable 5.2 qualifies. Try the model in the AI Gateway playground, and switch to the paid tier if it is unavailable to you.
Fact checked on 10 Nov 2026 against vercel.com [1], vercel.com [2]
When a client requires Zero Data Retention, this model returns an error through AI Gateway. Choose another model for that project.
Fact checked on 10 Nov 2026 against vercel.com [1]
Sources
- https://vercel.com/changelog/claude-fable-5-2-now-available-on-ai-gateway (official update)
- https://vercel.com/docs/ai-gateway/pricing
- https://vercel.com/ai-gateway/models
- https://v0.app/changelog
- https://v0.app/pricing
- https://vercel.com/ai-gateway/models/claude-fable-5.1
Facts in this article were checked against these sources on 10 Nov 2026. Product details change, so check the official page before you rely on them.