You can run Hermes Fable 5 setups three ways: with an Anthropic API key, through OpenRouter, or on your Claude subscription through the Claude Subscription DirectSDK plugin. Claude Fable 5 is Anthropic's model with a 1M-token context window and adaptive thinking that is always on. Hermes Agent, the open-source agent from Nous Research, can use it on any of those routes, and a set of plugin commits on 2 and 3 November 2026 changed how the subscription route handles Fable and Mythos.
📺 Watch: Building with Claude Fable 5 + Agent OS
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
This guide covers each route, what changed this week, what always-on thinking means inside Hermes, and how to set it up with hermes model.
Hermes Fable 5: what the model is
Anthropic's model page lists Claude Fable 5 with the model ID claude-fable-5, released on 9 June 2026. It has a 1M-token context window, up to 128K output tokens, text and image input, and a January 2026 knowledge cutoff. Thinking is adaptive and always on, with a default effort of high.
One thing to know before you start: Anthropic now marks Fable 5 as legacy. Claude Fable 5.1 is the current Fable model, and Anthropic recommends migrating. Fable 5 stays available, with retirement no sooner than 9 June 2027. Everything on this page applies to Fable 5.1 as well, and in places 5.1 is the easier choice.
Want the Hermes and Claude setups members are running right now? Join the AI Profit Boardroom ($69/mo). Or book a free AI strategy session to plan your own.
Three routes to run Fable 5 in Hermes
| Route | How you pay | Fable model ID | Thinking control |
|---|---|---|---|
| Anthropic provider with an API key | Pay per token on your Anthropic API organisation | claude-fable-5 | Hermes sets effort directly |
| OpenRouter | Your OpenRouter account | anthropic/claude-fable-5 | Hermes maps reasoning effort to OpenRouter's verbosity setting |
| Claude Subscription DirectSDK plugin (Experimental) | Your Claude plan, under Anthropic's plan rule for Fable | Listed when your account's picker offers it; the fable alias resolves to Fable 5.1 | Reasoning-off requests become low effort |
Route 1: Anthropic API key
The built-in Anthropic provider talks to the Claude API directly. With an API key, requests are billed per token to that key's organisation, independent of any Claude subscription. Hermes also supports Anthropic OAuth through hermes model, but its docs say that path only works on a Claude Max plan with extra usage credits added, and that Claude Pro subscribers cannot use it. For most people the API key is the simple choice here. Our Hermes and Claude guide covers the provider in full.
Route 2: OpenRouter
OpenRouter lists anthropic/claude-fable-5 with a 1M context window, alongside Fable 5.1. The Hermes docs note that adaptive-thinking models, including the Fable and Mythos class, ignore the usual reasoning effort field on OpenRouter. Hermes handles this for you: it sends your reasoning_effort as OpenRouter's verbosity parameter, which maps to Anthropic's effort setting. Setting effort to none leaves the model on its own adaptive default. You can also pin a Fable model to Anthropic as the only upstream with provider routing rules in config.yaml.
Route 3: your Claude subscription through DirectSDK
The Claude Subscription DirectSDK plugin is published by Nous Research in hermes-official-plugins and is labelled Experimental. Its README calls it a review build, not a production-readiness claim. It drives the official Claude Code CLI as the model client for your Claude subscription, while Hermes keeps its own agent loop, tools, approvals and compaction. It needs Hermes 0.21.4 or newer and a logged-in Claude Code install.
📺 Watch: Claude Fable 5 + Mythos is INSANE!
What the 2-3 November commits changed for Hermes Fable 5
Four commits landed in the DirectSDK plugin between 2 November 18:40 UTC and 3 November 00:24 UTC. Here is what each one means for you.
- Your own model list. The model picker now lists the models your account actually has, read from Claude Code's served picker. Older models your account still runs, such as Fable 5, appear when Claude Code offers them. Before this, the picker showed a generic list.
- The plan rule covers Fable 5 and 5.1. The commit quotes Anthropic's rule: "Pro bills both to usage credits, Max and Ultra include them." The picker shows a usage credits note on Pro. Team and Enterprise rows read "usage credits on standard seats", because the handshake names the plan but not the seat tier, and premium seats include Fable.
- Mythos 5.1 is access-gated. Anthropic offers Mythos 5.1 only to organisations it verifies, and describes it as the same model as Fable 5.1. The plugin now pins it with a 1M window and always-on thinking, but lists it only when your account's own picker offers it. It never appears from the built-in catalog and is never used as a fallback.
- The whole Mythos family always thinks. The mandatory-thinking rule now covers every claude-mythos model, so Mythos 5 no longer receives a thinking-disabled request, which Mythos rejects.
Not sure which plan or route fits your workload? Get a free AI strategy session.
📺 Watch: Claude 5 + Hermes Just Changed AI Agents Forever
Always-on thinking: what it means inside Hermes
Fable 5 cannot switch thinking off. Anthropic lists its thinking as adaptive and always on, so the model decides how much to think, steered by an effort level. That changes how a few Hermes settings behave.
- Reasoning off is not off. On the DirectSDK route, a reasoning-off request on a route that cannot disable thinking, Fable included, is sent as low effort instead. Expect some thinking on every turn.
- Effort is your main dial. Hermes accepts reasoning_effort levels from none up to ultra, with medium as its default. Each route translates that to what Anthropic accepts, clamping where needed.
- Pin effort per job. A Hermes cron job can pin its own reasoning effort. Routine jobs on Fable 5 can run at low effort while your main chat stays higher.
- Thinking costs context. On long agent loops, thinking and tool output fill the window faster. Hermes owns compaction on every route, including DirectSDK, where native autocompaction is turned off. See Hermes compaction for the thresholds.
Fable 5 context window in Hermes, per route
Anthropic lists 1M tokens for Fable 5, and OpenRouter lists the same. On the DirectSDK route, the README's model table pins Fable 5.1 at 1,000,000 tokens through the route claude-fable-5-1[1m], and the short name fable resolves to it. Fable 5 is not in that pinned table. It shows up when your account's picker offers it, and a model the plugin has not pinned keeps the label the CLI announced with an unpinned note. The README says a plain unpinned ID reports the 200K window that Claude Code applies to it, unless the CLI itself offers a [1m] form. If you need the full 1M window on your subscription, pick Fable 5.1.
Set up Hermes Fable 5 with hermes model
- Update Hermes. The latest release, v0.22.7, shipped on 3 November 2026.
- For the API route, set your Anthropic API key, run hermes model, choose Anthropic and pick claude-fable-5.
- For OpenRouter, run hermes model, choose OpenRouter, enter your key and pick anthropic/claude-fable-5.
- For your subscription, install Claude Code and log in with claude auth login. Then run hermes plugins install claude-subscription-directsdk.
- Run hermes model and pick Claude Subscription DirectSDK (Experimental). The flow checks that Claude Code is installed and logged in, then lists your account's models. Pick Fable 5 if it is offered, or Fable 5.1.
- Read the usage credits note on the row before you commit. On Pro, Fable bills to usage credits. If you do not want overage billing, the README says to turn off extra usage in your Claude account.
Existing installs may need Refresh models in the picker to clear an old cached list.
Fable 5 vs other Claude models in Hermes
Fable 5 is not the only Claude model you can run this way. Opus 5.5 and Sonnet 5.5 also have 1M windows on the DirectSDK route. For a head-to-head, read Claude Opus 5 vs Fable 5, and for the Opus setup see Hermes with Claude Opus 5.5. Julian's own model testing lives at Goldie Bench. To see how a model choice fits a full agent system, read the Agent OS guide.
Members of the AI Profit Boardroom get walkthroughs for each provider setup.
Hermes Fable 5 FAQ
Can I use Fable 5 in Hermes on a Claude Pro plan?
Yes, through the DirectSDK plugin, but Anthropic's plan rule bills Fable 5 and 5.1 to usage credits on Pro. Max and Ultra include them. Hermes' built-in Anthropic OAuth path does not support Pro.
Can I turn off thinking for Fable 5?
No. Thinking is always on. Lower the effort level instead; on DirectSDK a reasoning-off request is sent as low effort.
Should I use Fable 5 or Fable 5.1?
Anthropic marks Fable 5 as legacy and Fable 5.1 as current. On the DirectSDK route, Fable 5.1 is the pinned 1M route.
Why don't I see Mythos 5.1 in the picker?
Mythos 5.1 is only for organisations Anthropic has verified, and the plugin lists it only when your account's picker offers it. Need help choosing a route? Book a free AI strategy session.
