Claude Fable 5.1 Is Out: Anthropic's Frontier Model, Now in Runable's Ultra Mode

Sep 4, 2026•by Runable

Summarize with:

Claude Fable 5.1 Is Out: Anthropic's Frontier Model, Now in Runable's Ultra Mode

Key Takeaways

Anthropic released Claude Fable 5.1 on September 1, 2026, alongside Claude Mythos 5.1. It is Anthropic's most capable model for coding and long-running knowledge work.
Fable 5.1 roughly doubles Fable 5's score on Terminal-Bench-Science, and Anthropic cut cache read pricing by 75%, which lowers typical workload costs by about 25% and highly agentic workloads by up to 45%.
You can use Claude Fable 5.1 on Runable today: select Ultra in the model selector and the agent runs on Fable 5.1 behind the scenes.
Early adopters describe it as "friendly Fable": Fable-level intelligence at Opus-level price and Sonnet-level speed, about twice as fast as Opus 5 while using half the tokens.

Anthropic opened September with its biggest model release of the year. Claude Fable 5.1, launched on September 1, 2026 together with its research-focused sibling Claude Mythos 5.1, is the follow-up to a Fable 5 launch that had one of the rockiest starts in recent AI history. This time the story is different: a stable rollout, sharply better benchmarks, a 75% cut to cache read pricing, and availability everywhere from the Claude API to the major clouds.

It is also available on Runable as of this week. Flip the model selector to Ultra and your agent plans, codes, and researches with Fable 5.1 underneath. Here is what shipped, why it matters, and how to put it to work.

The road to Fable 5.1 was not smooth

Some context makes this release more impressive. Claude Fable 5 introduced Anthropic's fifth model generation on June 9, 2026, built for days-long, complex, asynchronous tasks that earlier models could not sustain. Three days later, on June 12, access went dark, with Anthropic apologizing for the disruption while it worked to restore service. Access came back on July 1, and in August the company updated Fable 5's biology safeguards to substantially reduce false positives.

Fable 5.1 is the payoff for that turbulence. It keeps the fifth-generation ambition, long-horizon autonomous work, and wraps it in a model that is faster, cheaper to run at scale, and noticeably more disciplined about how it works.

What is new in Claude Fable 5.1

Anthropic positions Fable 5.1 as a Mythos-level model built for your most ambitious, long-running projects. Three behaviors define it.

First, it avoids easy-seeming shortcuts. Where previous models would patch a symptom to make a test pass, Fable 5.1 is trained to fix root causes. Second, it sustains multi-day autonomous sessions: it plans the work, picks the tools it needs, recovers when a step fails, and keeps you updated without being asked. Third, it checks its own output, writing tests for its code and using vision to compare what it built against the original design or goal.

On measurable performance, the headline is science and agentic work. Fable 5.1 roughly doubled Fable 5's score on Terminal-Bench-Science 0.1, the benchmark that tests whether an agent can run real research workflows through code and terminal tools: analyzing data, running simulations, fitting models. For reference, the public leaderboard put Fable 5 in the low twenties, and OpenAI's own comparison charts two days later showed Fable 5.1 at 52.6%.

Vision improved too. Fable 5.1 reads diagrams, charts, and tables nested inside files and PDFs, which directly upgrades document-heavy work in finance, legal, analytics, and architecture.

Pricing: the cache read cut is the real story

Fable 5.1 is priced at $10 per million input tokens and $50 per million output tokens, the same premium tier as Fable 5. The change that matters is underneath: cache reads now cost $0.25 per million tokens, down 75% from Fable 5's $1.00.

Agentic workloads live and die on cache reads, because an agent re-reads its context every step of a long session. Anthropic estimates the cut reduces typical workload costs by about 25% and highly agentic workloads by up to approximately 45%. Cognition, which is moving its Opus 5 traffic in Devin to Fable 5.1, said the new cache pricing finally makes a Fable-class model economical for workloads it had kept on cheaper models, starting with code review.

For teams that need it, US-only inference is available at 1.1x pricing. Using Fable requires 30-day data retention for safety monitoring by default, with zero-data-retention options for eligible enterprise customers.

What early users are saying

The launch-day testimonials are unusually specific. Jane Street reports Fable 5.1 solves more of its coding problems than Fable 5 or Opus 5, achieves state of the art on trading intuition, and stays readable over long multi-step tasks where prior models became hard to follow.

A senior portfolio manager at Millennium describes a crash that occurred roughly once in a million runs and had gone unexplained for four to five years. Every model he tried missed it. Fable 5.1 disassembled an external vendor library, matched it against the core dump, and traced the crash to a bug in that library.

A MongoDB staff engineer had Fable 5.1 build a complex prototype in about three days, researching the company's services code first, then running for hours unattended with verification loops. He woke up each morning to the next phase finished, with a visual walkthrough and evidence it worked.

And the summary quote making the rounds: "It's friendly Fable. Fable-level intelligence, Opus-level price, Sonnet-speed." In those tests it ran about twice as fast as Opus 5 on half the tokens.

Safeguards and the Mythos 5.1 sibling

Fable 5.1 ships with robust safeguards for cybersecurity and biology. Queries in these domains that trip the safeguards are automatically rerouted to less capable models, cybersecurity tasks to Claude Opus 4.8 and biology tasks to Claude Opus 5, and you are not charged Fable prices for rerouted requests.

Claude Mythos 5.1, released the same day, is the specialized track: a model aimed at cyberdefenders and life scientists whose legitimate work requires the capabilities the safeguards would otherwise gate.

How to use Claude Fable 5.1 on Runable

Runable added Fable 5.1 to Ultra mode, the top tier of the model selector, above the Lite, Pro, and Max model tiers. Getting it running takes three steps.

Open a new task and click the model selector in the input bar, then choose Ultra. Higher tiers use more credits per task and reason better; Ultra now points the agent at frontier models, including Fable 5.1.

Describe the job in plain language: a full-stack app, a refactor across a codebase, a financial model, a research report. Fable 5.1's long-horizon autonomy is exactly what Runable's sandboxed agent workflow is built around, so multi-hour builds with verification loops are the sweet spot.

Refine in the same chat. Plan Mode, skills, connectors, memory, rollback, and branching all behave identically across model tiers, so switching to Ultra changes the brain, not the workflow. Runable has a free tier to start, with Pro at $20/mo as the most popular plan and Max at $100/mo for heavy usage.

Worth noting: OpenAI shipped GPT-6 Astra on September 3, two days after Fable 5.1, and it is also available in Runable's Ultra mode. A practical pattern is to fork the same task and run both, then keep the better result. Fable 5.1's reputation is stamina and root-cause discipline on deep coding work; Astra's edge is computer use and polished professional documents.

FAQ

What is Claude Fable 5.1?

Claude Fable 5.1 is Anthropic's flagship model, released September 1, 2026 alongside Claude Mythos 5.1. It is built for ambitious coding and knowledge work: multi-day autonomous sessions, root-cause debugging, self-testing code, and research workflows, with roughly double Fable 5's score on Terminal-Bench-Science.

How much does Fable 5.1 cost in the API?

Pricing is $10 per million input tokens and $50 per million output tokens, with cache reads at $0.25 per million, a 75% cut from Fable 5. Anthropic estimates typical workloads get about 25% cheaper and highly agentic workloads up to roughly 45% cheaper.

How do I use Fable 5.1 on Runable?

Select Ultra in the model selector in the input bar before sending your task. Ultra sits above the Lite, Pro, and Max tiers and now runs frontier models including Claude Fable 5.1. Everything else, from Plan Mode to rollback, works exactly the same.

Is Fable 5.1 faster than Opus 5?

Yes, according to early adopters. In customer testing it ran about twice as fast as Opus 5 while using roughly half as many tokens, which is why it is being described as Fable-level intelligence at Opus-level price and Sonnet-level speed.

What is Claude Mythos 5.1?

Mythos 5.1 is Fable 5.1's specialized sibling, released the same day for cyberdefenders and life scientists. It provides capabilities that Fable's safety systems gate for general users, offered through controlled access for organizations with legitimate defensive and scientific needs.

Run Claude Fable 5.1 in Runable's Ultra mode

The best test of a long-horizon model is a long-horizon task. Open Runable, set the model selector to Ultra, and hand Fable 5.1 something with real moving parts: an app with auth and payments, a codebase-wide refactor, a full research report. It starts free, and Ultra is one click away.