Claude Fable 5.1 review: what Anthropic's top model builds when nobody fixes its code

What is Claude Fable 5.1?

Anthropic's flagship generally available model, released on September 1, 2026. Anthropic describes Fable 5.1 and Claude Mythos 5.1 as "the same model, but with different levels of safeguards": Fable is generally available, Mythos only through trusted access programmes, according to its announcement. On the API it costs $10 per million input tokens and $50 per million output tokens, with cached reads at $0.25, and offers a 1-million-token context window with up to 128,000 output tokens, per Anthropic's model overview. Anthropic positions it "for demanding reasoning and long-horizon agentic work".

Anthropic's own headline numbers include 55.8% on Terminal-Bench 4.0 and 52.6% on Terminal-Bench-Science, more than double Fable 5's score on the latter. On the public Terminal-Bench leaderboard Fable 5.1 sits at 57.9%, behind Claude Opus 5.5 at 64.8%.

What did Claude Fable 5.1 build?

Eight projects, one prompt each, in Claude Code at maximum effort, with no human edits.

1. Landing page. One large Instrument Serif headline, "Get booked without giving up your best hours", over a soft gradient, with an animated cursor that books and pays for a session. 28 minutes.
2. Analytics dashboard. Correct figures, a best-day callout and a chart that flips into a table. 39 minutes.
3. Physics simulation. Big glossy numbered balls in a glowing hexagon, in the smallest file of any build on the site, 15 KB. 25 minutes.
4. 3D shooter. Duskfall, a rooftop arena under a violet sky; the player survives our whole 40-second script. 75 minutes, nine files, after a first attempt was stopped by a usage limit.
5. Phone finance app. Money-in and money-out chips and a "Sent!" overlay with the balance counting down. 35 minutes.
6. One-lane MOBA. Riftline, a bright low-poly pine forest with a river crossing, a glowing base platform, a shop, gold and creep-score counters, and a hero whose abilities run Arc Bolt, Frost Nova, Blink Step and Starfall. The hero holds the lane until the scripted push walks it into an enemy tower at 56 seconds. 131 minutes, fourteen files, on a fourth attempt after three were interrupted by infrastructure.
7. Open brief. Yonder, a $240 lamp that glows with the colour of the sky over a faraway city, on a dusky, quiet page. 26 minutes.
8. Kanban board. The most compact board of the four, with red-outlined overdue cards and a delete option in the details panel. 45 minutes, after a first attempt was cut short by an API outage.

How did Claude Fable 5.1 work?

In fewer, larger steps than Opus. Across its eight builds Fable used 41 to 67 turns per build and 9 to 44 shell commands, against up to 162 turns and 118 commands for Opus 5.5. It read about 44 million input tokens in total, most of them cheap cached reads at $0.25 per million, and wrote about 1.3 million output tokens at $50 per million, which is where most of its $111 of API-equivalent cost went. On the open brief it ran just nine shell commands: it wrote the page, looked, adjusted a little and stopped.

The exception is the MOBA, the most complex brief on the site, where Fable worked for 131 minutes, wrote fourteen script files and read 16 million input tokens. Even there it finished inside the limit, which Opus 5.5 did not manage on the shooter.

What is Claude Fable 5.1 good at?

Taste and economy of means. Fable's pages use fewer elements and more space than Opus's, and its choices hold together: one typeface pairing, one accent colour, one idea per screen. It is also efficient by Claude standards. Its physics simulation is 15 KB against 31 KB for Opus, and it used 41 to 67 turns per build where Opus used 81 to 162 on the briefs both finished. Where a brief invited a story, as the open brief did, Fable wrote a gentle one and built the page around it.

It is reliable too. Every Fable build that finished logged no runtime errors in our recordings, and its shooter's player survives our whole scripted input, which Opus 5.5's does not.

Where does Claude Fable 5.1 fall short?

Value against Opus 5.5, and patience with interruptions. Fable costs two and a half times as much per token as Opus, and because Opus works longer the two end up costing about the same per build: $8.75 to $24.76 for Fable at list prices, about $111 for all eight, against $7.15 to $21.06 for Opus on the briefs it finished cleanly. For that price Opus usually built more. Fable's restraint is a style, and when the brief calls for density, such as a dashboard full of controls or a game full of effects, the style shows as less.

Fable's long runs were also the ones most exposed to infrastructure trouble. Three of its eight briefs needed more than one attempt: a usage limit, an API outage and, for the MOBA, a sequence of both plus a stopped harness. None of these were model failures, and every interrupted attempt is listed on its build, but long agent sessions on a subscription meet limits sooner.

How does Claude Fable 5.1 compare with Claude Opus 5.5?

Quieter, faster, and usually less. On the three briefs both finished early on, Fable took 25 to 39 minutes where Opus took 40 to 81. On the landing page their designs are two different brands; on the dashboard they converged on the same typefaces and warm accents. Anthropic's own Opus 5.5 announcement says Opus 5.5 performs at Fable 5.1's level on most work at a lower price, and our builds agree.

How does Claude Fable 5.1 compare with GPT-6 Astra?

Same price per token, very different results. Both cost $10 and $50 per million tokens. Astra finished every brief in 8 to 25 minutes for $2 to $7 at list prices; Fable took 25 minutes to more than two hours for $9 to $25. Fable's builds are more distinctive and more polished; Astra's are tidy, complete and close to what the cheaper GPT-6.1 Sol produced.

Who should use Claude Fable 5.1?

Anyone whose work rewards restraint or long, careful reasoning. For building, it suits editorial sites, calm interfaces and briefs where less is more. For heavy, everyday building, Claude Opus 5.5 gives you more for the same money. And if you do pick Fable for a long build, run it on its own: its longest sessions were the ones most often cut short by usage limits. Note the plan detail before you count on it: Claude's pricing page lists Fable on Pro only through extra usage credits, and on the Max plans for up to half of the weekly limits.

Frequently asked questions

What is Claude Fable 5.1?

Anthropic's most capable generally available model, released on September 1, 2026 alongside Claude Mythos 5.1, which is the same model with different safeguards and limited access. Fable 5.1 costs $10 per million input tokens and $50 per million output tokens and is built for demanding reasoning and long-horizon agentic work.

Is Claude Fable 5.1 better than Claude Opus 5.5?

Not on our builds. Opus 5.5 produced richer work on most briefs and leads the public Terminal-Bench 4.0 leaderboard, at a lower token price. Fable 5.1 was faster, more restrained, and the better choice when you want a calm, editorial result.

Is Claude Fable free?

No. Claude's pricing page lists Fable as unavailable on the Free plan, available on Pro only through extra usage credits, and available on the Max plans for up to half of the weekly usage limits.

What happened to Claude Fable 5?

It was succeeded by Fable 5.1 on September 1, 2026. Anthropic says Fable 5.1 costs an estimated 25% less than Fable 5 for typical workloads, and up to about 45% less for highly agentic work, mainly through cheaper cached reads.