Fable 5 → API-only: please reconsider a subscription path (👍 + comment here if it affects you)

Status Fixed / completed

Maintainer reply None cached

Activity 7 comments · opened Jul 17, 2026 · closed Aug 28, 2026

If Fable 5's move to API-only affects your workflow, please 👍 this issue and add your use case in a comment. I'm consolidating the scattered same-day reports into one properly-worded thread so the signal is easy to read.

What changed

As of 2026-07-17, selecting Fable 5 (/model → Fable 5) and issuing any request on a Claude Max subscription returns:

Anthropic API Error: Usage credits are required for this model.

Fable 5 still appears in the model picker, but it is no longer usable under the included subscription allowance — it now requires separate API / credit billing. Observed first-hand on Claude Code v2.1.212 (Linux, Max plan), including on /compact.

The move looks announced — the timing looks early

My understanding is the API-only transition is a deliberate, announced change, not a pure billing glitch. A commenter here notes Anthropic communicated a 19 July date for it — so its arrival today (17 July) looks like a premature/early rollout, which is why several same-day reports read it as a bug:

#78610 — Incorrectly Prompts for API Credits Despite Active Subscription Usage Limit
#78611 — Anthropic API Error: Usage credits are required for this model

Both things are true at once: the transition is real and scheduled, and it appears to have fired two days early. Either way, the request here is a product reconsideration, not just a timing patch. (This is also distinct from the older "Usage credits required for 1M context" cluster, which is a separate long-context trigger.)

Why this is a painful change

Fable 5 filled a genuinely useful niche: capable enough for agentic and mechanical work — parallel dispatch, sidebar tasks, mechanical edits, judgment-light passes — yet cheap enough to run at volume inside a subscription. Removing it from the included allowance means that class of work now either:

requires separate API credits on top of an active subscription, or
gets pushed onto Opus / Sonnet — heavier and more expensive for work that never needed them, which worsens the effective cost for exactly the high-volume, multi-agent Claude Code users who adopted Fable for this reason (see also #77964 on cost-efficient model routing).

Fable is good. That's why losing it from the subscription stings rather than shrugs.

Market context

The included cheap-but-capable tier is becoming a baseline expectation across the field, not a Claude-specific nicety — OpenAI's GPT‑5.6 (GA July 2026), for one, keeps Sol and Terra (both in Fable's capability class) selectable on a ChatGPT Plus/Pro subscription. Moving Fable the other way — behind metered API billing — cuts against that direction. I raise this as market context, not leverage: I build on Claude Code and intend to keep doing so — which is exactly why I'd rather the cheap-capable tier stayed in the subscription.

The ask — any one of these resolves it

Restore Fable 5 to the included Max/Pro subscription allowance, or
Offer a capped included Fable allowance — even a modest cap restores the workflow, or
If the API-only move is firm, publish the rationale and roadmap so users can plan, and consider a subscription-tier "cheap, capable" model to fill the niche Fable is vacating.

Environment

| | |
|---|---|
| Claude Code | v2.1.212 |
| Platform | Linux |
| Plan | Claude Max |
| Date observed | 2026-07-17 |

Also submitted via in-CLI /feedback. Happy to provide more detail or logs.

Thank you for building Claude Code — this is raised in that spirit.

View original on GitHub ↗

6 Comments

davidvgalbraith · 1 month ago

The bug reports are because Anthropic's prior communication stated this transition would happen on 7/19, so it happening on 7/17 is a likely bug. +1 to having it never happen though!

sbp · 1 month ago

This was a bug, not deliberate:

Monitoring - We have applied a fix to resolve the erroneous requirement for usage credits on Fable 5, and are monitoring to ensure no further issues. You may need to relaunch Claude.ai or Claude Code to restore access as expected.

Jul 17, 2026 - 18:48 UTC

https://status.claude.com/

So this issue is invalid.

powell-clark · 1 month ago

Update — confirmed working again here: Fable 5 on a Max subscription, Claude Code (Linux), no usage-credit error after relaunch. Matches the status-page fix (Jul 17, 18:48 UTC): today's credits wall was a bug, not the announced transition.

Two things follow:

If Fable is restored for you too, a quick 👍 or comment here helps confirm the fix landed for everyone, not just some accounts.
The original ask stands for the announced July 19 API-only date (screenshot upthread). If that transition proceeds as communicated, everything in the issue body applies from that date — so keeping this open until the 19th clarifies.

Related reports from today's incident: #78628 #78633 #78634 #78639.

powell-clark · 1 month ago

Edit note: I've softened the earlier wording. Full transparency — I draft these with an AI assistant, and the strong competitive line was Opus's phrasing; it came out reading more like leverage than I meant, so I put a human hand back on the dial. For the record: I build on Claude Code and I'm staying — the GPT‑5.6 mention is market context, not a threat to walk. The ask is unchanged: keep Fable's cheap-but-capable tier available on subscription. (Some irony in Opus over-egging a plea to save Fable.) Thanks to everyone weighing in.

heytcass · 1 month ago

Nope. Hasn't landed for me. In fact, it's worse. It burnt through $30 of usage credits while I was on Opus, after they claimed the fix was in place.

WMSmith25 · 1 month ago

As someone who’s maintained active subscriptions to every major AI platform since ChatGPT’s launch, I’ve watched the space evolve—and the hype around Fable 5 doesn’t match the reality. Let’s be clear: Fable 5 is not budget-friendly. Its planned pricing places it in a tier that’s hard to justify when competitors offer just as capable (and in some cases, more capable) models at a fraction of the cost of having to pay for separate API spend outside of our subscription. All you will do is piss off developers who actually pay for the max plans to shift our higher tier memberships to the numerous other platform premium tier offerings to take its place.

Anthropic’s Miscalculation
Anthropic’s decision to price Fable 5 as a premium offering at the planned insane API cost only is a misstep for two reasons:

Developers who understand the tech know that Fable 5’s capabilities are not unique—they can get almost the same results from a properly specked and instructed agentic workflow using cheaper options from competitors like Mistral, Google, xAI, OpenAI, or self-hosted models for far less money if forced to the API only route.
The "too powerful to let loose" narrative is transparent. If you give a competent AI model the right tools and proper instruction, it will find vulnerabilities, debug code, and perform agentic tasks—Fable 5 is not special in this regard.

The result? Developers who know better will simply jump to Mistral, Google, xAI, OpenAI, or self-hosted alternatives—models that don’t overcharge for capabilities that are now table stakes.
The Bottom Line
Anthropic is betting on short lived security theater on a close source model that restricts model usage for common tasks while charging a premium price in a market that’s rapidly moving toward accessibility, openness, and cost efficiency. With Gemma 4 and Mistral’s latest models offering near equal performance on the overwhelming majority of common tasks at a fraction of the cost, Anthropic risks losing the very users who actually push the boundaries of what these models can do—the ones who won’t pay a premium for restrictions disguised as features.

Showing cached comments. Read the full discussion on GitHub ↗