Claude Fable 5

Claude Fable 5 launched on Jun 9, 2026 as Anthropic's first Mythos-class model for general use, was suspended for all users on Jun 12 when US export controls applied to it, and returned on Jul 1, 2026. Claude Fable 5.1 replaced it as the current model in Sep 2026; Fable 5 remains on the API as a legacy model.

Claude Fable 5 was released on Jun 9, 2026 as "a Mythos-class model" that anyone could use, priced at $10 per million input tokens and $50 per million output tokens, and three days later Anthropic suspended it for every user after the US government applied export controls to it (Redeploying Fable 5, Anthropic, Jun 30, 2026). It came back on Jul 1, 2026, with a new classifier in front of it, and was succeeded in Sep 2026 by Claude Fable 5.1. Fable 5 is still served on the API as a legacy model, as of Nov 6, 2026.

Quick facts

Released Jun 9, 2026
Suspended Jun 12, 2026, for all users
Redeployed Jul 1, 2026
API id claude-fable-5 (Bedrock anthropic.claude-fable-5)
Price $10 input, $50 output per million tokens; cache reads $1
Context window 1M tokens (500K in Claude chat)
Max output 128K tokens
Knowledge cutoff Jan 2026
Status Legacy, still available (as of Nov 6, 2026)
Retirement Not before Jun 9, 2027
Successor Claude Fable 5.1

Sources: Claude Fable 5, Pricing and Model deprecations, Anthropic docs, read Nov 6, 2026; How large is the context window on paid Claude plans?, Claude Help Center, read Nov 6, 2026.

What launched on Jun 9, 2026

Anthropic announced two names for one underlying model. Claude Fable 5 was the general release; Claude Mythos 5 was the same model with some safeguards lifted, at first only for Project Glasswing partners doing defensive security and soon for selected biology researchers (Claude Fable 5 and Claude Mythos 5, Jun 9, 2026). Anthropic called Fable 5 "less than half the price" of the restricted Claude Mythos Preview, and said Mythos 5 was "comparable to, or somewhat stronger than" that preview in most cases.

Fable 5 went to the Claude API, the major clouds and consumption-based Enterprise plans. Pro, Max, Team and seat-based Enterprise subscribers could use it at no extra cost through Jun 22; from Jun 23 their use needed usage credits. All Mythos-class traffic was kept for 30 days on every surface and, Anthropic said, not used for training. When its safeguards flagged a topic, Fable 5 handed the request to Claude Opus 4.8, and Anthropic said more than 95% of Fable sessions involved no such fallback (Claude Fable 5 and Claude Mythos 5).

The launch post's benchmark table is an image. Its text claims the top score of any model on the Hebbia Finance Benchmark and the best result among frontier models on Cognition's FrontierCode, even at medium effort. On Terminal-Bench 4.0's public leaderboard, Fable 5 at max effort reached 44.55% (row dated Jun 9, 2026, read Nov 6, 2026).

A bar chart titled Agentic coding, success rate in percent. On SWE-Bench Pro, Claude Fable (orange) scores 80.3, Claude Opus 4.8 (green) 69.2 and GPT 5.5 (gray) 58.6. On FrontierCode, Claude Fable scores 29.3, Claude Opus 4.8 13.4 and GPT 5.5 5.7.
Anthropic's agentic coding chart from the Fable 5 launch: Claude Fable scores 80.3 on SWE-Bench Pro and 29.3 on FrontierCode, ahead of Claude Opus 4.8 and GPT 5.5. Anthropic

Why was Claude Fable 5 suspended?

Anthropic's account, published on Jun 30, 2026 (Redeploying Fable 5):

  1. Before Jun 12, Amazon researchers reported a way around Fable 5's safeguards that got it to write code showing how a vulnerability could be exploited.
  2. On Friday, Jun 12, the US government applied export controls to Fable 5 and Mythos 5 that restricted access by foreign nationals. A US Executive Order of Jun 2, 2026 had set out the government's approach.
  3. Anthropic could not check every user's nationality in real time, so it suspended access to both models for all users, everywhere.

Anthropic's statement that evening said the directive arrived at 5:21pm ET, called it "a misunderstanding", and argued the jailbreak it cited was narrow and that other models, including OpenAI's GPT-5.5, could do the same (Statement on the directive to suspend Fable 5 access, Anthropic, Jun 12, 2026).

Anthropic then tested the reported technique and found that less capable models could reproduce the same findings and exploit demonstration. It trained a classifier that, it says, blocks the technique "in over 99% of cases" and sends blocked requests to Opus 4.8. The US Center for AI Standards and Innovation called the safeguards "extraordinarily strong", in Anthropic's account.

How Fable 5 came back

On Jun 26, the government approved Mythos 5 access for a set of US organizations, and on Tuesday, Jun 30, Anthropic reported that the export controls on both models had been lifted. Fable 5 returned worldwide on Wednesday, Jul 1, on the Claude Platform, Claude.ai, Claude Code and Cowork, with Amazon, Google Cloud and Microsoft Foundry to follow "as quickly as possible" (Redeploying Fable 5). The original launch post now carries dated notes for the Jun 12 suspension and the Jul 1 return (Claude Fable 5 and Claude Mythos 5, read Nov 6, 2026).

The allowance and the one-time credit

After the return, Fable 5 was included for up to 50% of the weekly usage limit on Pro, Max, Team and select Enterprise plans. Anthropic's Jun 30 post said this would run through Jul 7, 2026 (Redeploying Fable 5); the newer Help Center article says the promotion ended on Jul 19, 2026 at 11:59 PM PT and covered Fable 5 only (Claude Fable models on your plan, Claude Help Center, read Nov 6, 2026). The later date is the one Anthropic's support pages now give.

When Fable 5 moved to usage credits, Anthropic gave a one-time credit: $100 of usage credits for each qualifying Pro plan, and $100 per purchased Team standard seat up to $2,500 per organization. Max and Enterprise were not eligible, nor were free trials. The credits could be claimed from Jul 20 through Aug 2, 2026, worked on any model, and expired on Sep 17, 2026 (Claude Fable 5 one-time free credits promotion, Claude Help Center, read Nov 6, 2026). There is no such credit for Fable 5.1.

How Fable 5.1 replaced it

Claude Fable 5.1, released on Sep 1, 2026 (Anthropic's newsroom lists it on Sep 15), kept the $10 and $50 token prices, cut cache reads from $1 to $0.25 per million, and loosened the cyber and biology safeguards that had made Fable 5 refuse or fall back more often (Introducing Claude Mythos and Fable 5.1, Sep 15, 2026). In Anthropic's table, Fable 5.1 scored 55.8% on Terminal-Bench 4.0 against Fable 5's 42.0%. Meanwhile Opus 4.8 stayed the fallback model, and Claude Opus 5 (Jul 24, 2026) arrived at half Fable 5's price, which Anthropic described as close to Fable 5 (Introducing Claude Opus 5, Jul 24, 2026).

What reviewers found

Bijan Bowen, who made three videos testing Fable 5, said in a Sep 1, 2026 video that it "felt like a proper next generation model". He also recalled what bothered many users: a chat could end with a notice that the safeguards had stopped it and Opus 4.8 was available instead, and sometimes the chat could not go on at all. He welcomed Fable 5.1's looser safeguards for that reason.

Frequently asked questions

Is Claude Fable 5 still available? Yes. As of Nov 6, 2026, Claude Fable 5 is a legacy model still served on the Claude API as claude-fable-5, with retirement no sooner than Jun 9, 2027. In the apps, Claude Fable 5.1 is the current Fable model.

Why was Claude Fable 5 shut down in June 2026? The US government applied export controls to Claude Fable 5 and Mythos 5 on Jun 12, 2026, restricting access by foreign nationals. Anthropic could not verify nationality in real time, so it suspended both models for everyone until the controls were lifted on Jun 30.

How long was Claude Fable 5 offline? Claude Fable 5 was unavailable from Jun 12 to Jul 1, 2026, about 19 days, on Anthropic's own products; the cloud platforms were re-enabled after that.

What is the difference between Claude Fable 5 and Mythos 5? They are the same model. Claude Mythos 5 runs with some cyber or biology safeguards lifted and is limited to approved organizations; Claude Fable 5 is the general release.

Should I move from Claude Fable 5 to Fable 5.1? Anthropic recommends it: Claude Fable 5.1 costs the same per token, reads cached input at a quarter of the price and scores higher in Anthropic's tests. Check the breaking changes on the Fable 5.1 docs page first, such as the error on forced tool use.

Sources

Moments

Introducing Claude Fable 5

Anthropic

  1. Claude Fable 5 was released on Jun 9, 2026 as "a Mythos-class model" that anyone could use, priced at $10 per million…
  2. Fable 5 went to the Claude API, the major clouds and consumption-based Enterprise plans. Pro, Max, Team and seat-based…

Claude Fable 5: working at the frontier

Claude

  1. The launch post's benchmark table is an image. Its text claims the top score of any model on the Hebbia Finance…

Claude Fable 5.1 Is INSANE - Hands-On With the BEST Model Yet!

Bijan Bowen

  1. Bijan Bowen, who made three videos testing Fable 5, said in a Sep 1, 2026 video that it "felt like a proper next…

Photos

A bar chart titled Agentic coding, success rate in percent. On SWE-Bench Pro, Claude Fable (orange) scores 80.3, Claude Opus 4.8 (green) 69.2 and GPT 5.5 (gray) 58.6. On FrontierCode, Claude Fable scores 29.3, Claude Opus 4.8 13.4 and GPT 5.5 5.7.

See also

About this page

Last updated Nov 6, 2026 by the Vi.ki team. Page text CC BY-SA 4.0.

Edit History

Hotkeys

/
Jump to the search box
? or Ctrl + /
Open this list
Esc
Close this list, the photo viewer, or the account menu

Video

Space or K
Play or pause
← →
Back or forward 5 seconds
J L
Back or forward 10 seconds
0 to 9
Go to 0% to 90% of the video
, .
Previous or next frame, while paused
M
Mute or unmute
C
Captions on or off
F
Full screen on or off
A D
Previous or next moment

Photo viewer

← → or A D
Previous or next photo
↑ ↓ or W S
Previous or next photo
Home End
First or last photo
I
Show or hide the photo's info
Z
Clickless on or off: a pointer over a thumbnail shows it
N
Compare with the last photo: back to the one shown before this, and N again comes back
Q
Move the thumbnails to the next edge: top, bottom, left and round
, or Ctrl + ,
Open or close the viewer's settings
V
Clickless pan: a zoomed photo follows the pointer, rather than being dragged
Shift
Browse on or off
B
Show or hide the button bar
T
Show or hide the thumbnails
1
100% zoom on or off
0
Rate the photo as no opinion; again takes the answer back
+ -
Rate the photo a step higher or lower
F
Full screen: the photo alone. F again puts everything back.
Back to top