Hermes Fable 5 earns a 7.4 out of 10 from me overall: an excellent model behind a capable agent, held back by an experimental plugin and plan rules that punish Pro accounts.
That number is my editorial reading of the docs, not a benchmark.
I scored both setup routes, the official claude-subscription-directsdk plugin and Hermes's built-in Anthropic provider, against eight criteria.
Each score has a one-line reason, tied to the README, the commits or the vendor pages.
If you want a second opinion on your own setup, book a free SEO strategy session with my team.
How I Scored It
Every score below comes from reading the sources, not from running benchmarks.
I used the plugin README, the two commits from 2 November, the Hermes providers guide, the cron and delegation docs, and Anthropic's model and plan pages.
Each criterion is scored out of 10.
A 10 means the docs show nothing missing for that criterion.
A 5 means the docs themselves flag a real gap.
Where the two routes differ, I note which one the score applies to.
The Scorecard
| Criterion | Score | Why | |
|---|---|---|---|
| Model capability | 9/10 | Anthropic's [[overview | models overview]] reserves Fable 5.1 for demanding reasoning and long-horizon agentic work |
| Context window | 10/10 | Fable 5.1 has a 1M-token window, and the plugin pins it as claude-fable-5-1[1m] | |
| Plan fit on Max and Ultra | 9/10 | Commit [[c1 | fdbf4225d2]] marks Fable as included on both plans |
| Plan fit on Pro | 4/10 | The same commit bills Fable to usage credits on Pro from the first request | |
| Setup effort | 7/10 | Three commands plus a Claude Code login, but two moving parts to keep updated | |
| Maturity | 5/10 | The [[plugin | README]] calls the plugin a review build, not production-ready |
| Cost guard rails | 7/10 | /usage, per-job cron pins and a delegation model, but no account-level overage cap in the plugin | |
| Built-in route flexibility | 8/10 | API key, setup-token or Claude login, all on a core provider |
The eight scores average 7.4, but the two plan rows never apply to the same buyer.
For a Max or Ultra buyer it lands higher.
For a Pro buyer it lands lower.
Four Strengths The Docs Back Up
The strengths are about fit, not hype: each one traces to a line in the README, a commit or the Hermes code.
A 1M-Token Window On Both Routes
Fable 5.1 reads up to 1M tokens and writes up to 128K, per Anthropic's Fable 5.1 page, and its thinking never switches off.
The plugin pins Fable 5.1 to claude-fable-5-1[1m] with a 1,000,000-token window.
Hermes's own Anthropic adapter, in anthropic_adapter.py, also treats Fable as a mandatory-thinking model with a 1M window and 128K output.
So whichever route you pick, Hermes budgets context the way Anthropic documents it.
That matters for agents that read long files and keep long sessions open.
No API Key On The Plugin Route
The plugin drives the unmodified Claude Code CLI as a model provider.
You log in once with claude auth login, and Hermes rides that login.
The README says the plugin never opens, copies, refreshes or prints the credential files.
Hermes keeps its own agent loop, tools, approvals and compaction, so you lose none of the agent features.
Hermes Won't Put You On Fable By Accident
This is a quiet point in Hermes's favour.
The models.py code classifies Opus and Fable as frontier tiers that are unsafe as silent defaults.
A freshly connected provider never lands you on Fable without you choosing it.
That stops a frontier model from leaking into every cron job that doesn't set its own model.
🔥 Want the full Hermes stack scored and set up for you? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members comparing agent setups. → Join the AI Profit Boardroom
The Picker Shows What Your Account Can Run
Commit 4220bbaadb changed how the model list is built.
The plugin now asks Claude Code for the account's own served picker, without sending a user turn.
So entitlement-gated and older models appear exactly when your account can use them.
A pinned fallback list keeps older models such as Opus 4.8 selectable when the live picker drops them.
One catch: claude_code_telemetry is on by default, because it fetches the feature flags that unlock models in that picker.
Switch it off and the list can shrink to each family's current model.
Four Limits You Should Weigh
The limits are just as concrete, and two of them come straight from the plugin's own README.
It's An Experimental Plugin
The provider's own name ends in "(Experimental)".
The README is honest about the gaps.
It lists unsupported surfaces such as forced tool choice, assistant prefill and parallel_tool_calls=False.
It says replay behaviour depends on the Claude Code version, so a CLI update can change things.
That's why maturity scores a 5.
Pro And Standard Seats Pay Usage Credits
Anthropic's Fable plan article says Fable 5 and 5.1 are a standard part of Max, Ultra and premium seats.
On Pro and on standard Team and Enterprise seats, both run on pay-as-you-go usage credits.
The plugin's picker shows that with a dim "usage credits" note on the Fable row.
The README even records a Pro test account rejecting Fable 5.1 with "Fable 5.1 requires usage credits".
The Built-In OAuth Path Is Narrow
The built-in Anthropic provider offers an OAuth login through hermes model.
The Hermes providers guide says it only works on Claude Max with purchased extra usage credits.
Hermes spends only those extra credits on that path, never the base Max allowance.
Claude Pro can't use the OAuth path at all.
So the practical built-in options are an API key, billed per token to the key's organization, or Max plus extra credits.
No Hermes Page Says "Fable Support"
I looked for an official Hermes docs page or release note that names Fable support.
There isn't one.
Fable appears in the plugin README and its commits, in the Anthropic model list inside models_catalog_static.py, and in a provider-routing config example.
That's enough to run it, but it isn't a documented, versioned support promise.
What You See In /usage On Each Route
This is a practical difference between the two routes.
On the plugin route, /usage asks Claude Code for your plan limits, so you see where you stand against your subscription.
The README adds that the native cost figure Hermes records is a list-price estimate, not proof of what your plan charged.
On the built-in route, the Hermes tips guide says /usage shows your token consumption, and /insights gives a 30-day view.
Neither route adds an account-level cap of its own.
So the README's advice is to disable extra usage at the account level when overage billing is unwelcome.
That's the only hard stop, and it lives in your Claude account, not in Hermes.
Plugin Route vs Built-In Route, Head To Head
| Plugin route | Built-in Anthropic route | |
|---|---|---|
| Best for | Max and Ultra owners | API-key users and Max owners with extra credits |
| Login | Claude Code CLI | API key, setup-token or Anthropic OAuth |
| Fable 5.1 id | fable → claude-fable-5-1[1m] | claude-fable-5-1 |
| Maturity | Experimental | Core provider |
| Usage view | /usage asks Claude Code for plan limits | /usage shows token use and context |
| Winner | Plan fit | Stability |
My call: the plugin wins for subscription owners who want Fable's allowance working inside Hermes.
The built-in route wins for anyone who wants the most stable path, or billing kept apart from their Claude plan.
Who Should Run Hermes On Fable
A bookkeeper doing quarter-end reviews.
- Load a client's full quarter of records into one Fable session.
- Ask Hermes to reconcile them against the prior quarter's notes.
- Have it list every mismatch with the line it came from.
- Review the list and send the client a query.
A roofing contractor writing complex job scopes.
- Paste in the survey notes and the supplier spec sheets.
- Ask Fable to write a scope of works with the order of tasks.
- Have a lighter subagent turn the scope into a materials checklist.
- Check both before the site visit.
A recruitment agency shortlisting for a senior role.
- Drop the job spec and every application into one Fable session.
- Ask Hermes to score each application against the spec, criterion by criterion.
- Have it explain each score with the line from the application it relied on.
- A recruiter reads the shortlist and makes the calls.
Both are long, multistep reasoning jobs, which is exactly what Anthropic's model guide points Fable at.
Who Shouldn't
Skip it if your agent mostly runs short, repeatable jobs like inbox triage or daily summaries.
Anthropic itself says to start with Opus 5.5 for most workloads.
Skip the plugin if you need a production-grade path today.
Skip Fable on Pro if you don't want every request drawing usage credits.
And if you do run it, pin your cron jobs with hermes cron edit <job_id> --model <model> from the cron docs and set a lighter delegation.model from the delegation docs.
For the wider model shortlist, see my best Hermes Agent model verdict.
📞 Want a second pair of eyes before you commit? Book a free SEO strategy session and my team will look at which route and which models fit your business. → Book a free strategy call
Fable 5 vs Fable 5.1: Which One The Verdict Covers
This verdict is about Fable 5.1.
Anthropic released claude-fable-5-1 on 1 September 2026, and Fable 5 is now legacy.
The plugin commit history applies the same plan rule to both models, and the account picker lists Fable 5 too.
New setups should pick Fable 5.1, since the plugin's fable short name already resolves to it.
Final Verdict
Hermes on Fable is the strongest agent brain you can put behind Hermes from Anthropic's public lineup.
On Max or Ultra, run the plugin, set Fable as the main chat and push the rest to lighter models.
On Pro, start with the built-in route and an API key, or accept usage credits, and test one real job first.
Either way, treat the plugin as experimental and keep a person on the final step.
Here's the verdict by job type.
- Long research, audits and multi-document analysis: run Fable 5.1 as the main chat.
- Batch rewrites, summaries and checklists: hand them to a Sonnet or Haiku subagent.
- Scheduled jobs: pin them to a lighter model, never leave them following Fable.
- Client-facing production agents: prefer the built-in Anthropic route with an API key until the plugin leaves experimental status.
Want the whole agent system ready to install? Join the AI Profit Boardroom and get the Agent OS.
Also On Our Network
- 🌐 The exact commands and config for both Hermes-on-Fable setup routes
- 🌐 Which Claude plan covers Fable in Hermes, and four small-firm use cases
- 🌐 Five prompts that make your agent switch Hermes onto Claude Fable
- 🌐 Which Claude models to pair with Fable in Hermes, by GoldieBench score
- 🌐 The complete Hermes Fable 5 guide on the AI Profit Boardroom blog
FAQ: Hermes Fable 5 Verdict
What is the overall verdict on running Hermes with Fable?
Strong for Max and Ultra owners who want Fable as the main chat, weaker for Pro owners because every Fable request draws usage credits. The plugin is experimental, so keep human review in the loop.
Does an official Hermes docs page say Hermes supports Fable?
No Hermes docs page or release note names Fable support directly. Fable shows up in the official plugin README and commits, in Hermes model catalogs in the code, and in config examples.
Is the plugin route or the API-key route better?
The plugin route wins on plan fit for Max and Ultra. The built-in Anthropic route with an API key wins on maturity and on keeping billing separate from a Claude plan.
Who should not run Hermes on Fable?
Anyone whose agent jobs are short and repeatable. Anthropic suggests starting with Opus 5.5 for most work and moving to Fable 5.1 only when demanding reasoning or long agent runs need it.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Score your own plan against the table above, pick the route that wins for you, and book a free strategy session if you want my team to check the call before you commit to Hermes Fable 5.
