
Anthropic shipped Claude Fable 5 on Amazon Bedrock today, and Foxl v0.3.3 supports it on day one. Fable 5 is Anthropic's new state-of-the-art model - top-tier on nearly every benchmark, built for ambitious, long-running work with strong vision. It is selectable everywhere Foxl talks to a model: the Foxl App chat, the Foxl Code coding agents, and Foxl Notes summaries.
Opus 4.8 stays the default. Fable 5 is the model you reach for when you want the most capable option Foxl can give you - pick it from the model selector. It is a Pro/Ultra model, excluded from the free tier the same way the Opus tiers are. Pricing is $10 / $50 per million input / output tokens (cache reads $1, 5-minute cache writes $12.50).
One catalog entry, every surface
Foxl keeps a single model catalog as the source of truth. Adding Fable was one entry there, and it flows out to every provider, the relay credit meter, the tier gate, the usage charts, and all three model pickers automatically. Three transports light up at once:
- Foxl relay (your Foxl credits) - the default path for the web app and most desktop users. Metered against your credit balance at the rates above.
- Bring your own key - your Anthropic API key, no Foxl credits consumed.
- Claude Code (OAuth) - Fable 5 on your existing Claude Pro/Max subscription, right alongside Opus 4.8 / 4.7 and Sonnet 4.6 in compatibility mode.
The parts that were not a one-line model swap
A new model is rarely a clean drop-in, and Fable 5 had three sharp edges worth calling out - all now handled so you never see them.
It only ships a global inference profile. Unlike Opus and Sonnet, which expose both global. and us. cross-region profiles on Bedrock, Fable 5 only has global.anthropic.claude-fable-5 in the US - the us. form is simply invalid. Foxl routes both its desktop and relay paths to the same global profile.
It requires data-retention opt-in, per region. Fable 5 and its sibling Mythos 5 require the account to opt into provider data sharing before Bedrock will serve them; without it, every call returns "data retention mode 'default' is not available for this model". That opt-in is region-scoped, and Fable is GA in us-east-1 (and eu-north-1), not in the us-west-2 region our relay signs against by default. So Foxl pins Fable to us-east-1 on both the relay and the desktop direct-to-Bedrock path. Every other model keeps using its normal region - only Fable's traffic touches the retention boundary.
Its thinking shape is adaptive, not budgeted. Fable uses adaptive thinking with an effort level (thinking.type: "adaptive" + output_config.effort), and it rejects the older budget-token shape outright. Foxl now derives the thinking shape from the catalog rather than matching on model names, so Fable gets adaptive automatically - and the per-request task budget only rides on the API surfaces that accept it (the Anthropic Messages path), not Bedrock Converse, which rejects it.
A note on capacity
Fable 5 is brand new, and AWS is expanding access gradually based on usage. In the first hours after launch a meaningful share of calls came back as transient 503s ("Bedrock is unable to process your request") - an AWS-side capacity ramp, not a Foxl bug, and Opus/Sonnet were unaffected. We added a retry strategy that transparently retries those transient 5xx responses (not real validation errors), so the intermittent failures recover on their own. As AWS capacity expands, this fades entirely.
How to use it
Open the model selector in chat, in a Foxl Code task, or in Foxl Notes settings, and pick Fable 5. On the desktop you can also select it under Claude Code (OAuth) to run it on your Claude subscription. Everything else - tool use, adaptive thinking, streaming, prompt caching - flows through Foxl's normal agent loop exactly as it does for Opus.