Claude Fable 5 vs Sonnet 5: Which Claude 5 Model Should You Actually Use?

Anthropic shipped two very different members of the Claude 5 family within three weeks of each other: the top-tier Claude Fable 5 and the mainstream Claude Sonnet 5. The short answer is that Fable 5 is raw maximum power at a flagship price, while Sonnet 5 is the balanced, everyday workhorse most people should reach for first.

Both come from Anthropic, both sit inside the Claude 5 generation, yet they are separated by roughly a 5x gap in output pricing and a clear split in purpose. This guide breaks down what each model is, how they compare on capability, price, and safety, and which one fits your workload.

Meet the two models: Fable 5 and Sonnet 5

The two models were built for opposite ends of the same spectrum. One chases the absolute frontier of capability; the other chases the best value for daily agentic work.

What Claude Fable 5 is

Claude Fable 5 launched on June 9, 2026 as Anthropic’s most capable widely released model, a “Mythos-class” system that is state of the art on nearly every tested benchmark, from software engineering and knowledge work to vision and scientific research. Its defining trait is that the longer and more complex the task, the larger its lead over Anthropic’s other models grows. It carries the model id claude-fable-5 and shipped with senior-level reasoning results on finance benchmarks and strong long-context and vision performance.

What Claude Sonnet 5 is

Claude Sonnet 5 arrived on June 30, 2026 as the successor to Sonnet 4.6 and the most agentic Sonnet model yet. It plans, reasons, and uses tools like a browser and terminal, coordinates subagents, and runs autonomously. On launch day it became the default model across Claude’s Free and Pro plans, and it is also available to Max, Team, and Enterprise users, in Claude Code, and through the Claude Platform API under the id claude-sonnet-5.

Capability and benchmark comparison

Comparing these two is less about a single leaderboard number and more about matching horsepower to the job. Fable 5 wins on raw ceiling; Sonnet 5 wins on efficiency per dollar.

Raw power: where Fable 5 leads

Fable 5 is the stronger model in absolute terms. Anthropic describes it as state of the art on nearly all tested benchmarks and reports that it outperforms even the flagship Opus 4.8 across the evaluations it ran. Its advantage is most pronounced on long, autonomous tasks: deep research runs, repository-level engineering, and multi-step analysis where a model has to hold context and reason for extended stretches. Because the release is very recent, precise cross-model percentages for Fable 5 are not consistently published, but the qualitative picture is clear: for the hardest problems, Fable 5 sets the pace. Anthropic also points to concrete operational wins, including senior-level results on finance reasoning benchmarks, top marks on frontier coding evaluations, and vision skills strong enough to rebuild working web apps from a screenshot. In practice that ceiling shows up as fewer hand-offs: a single Fable 5 run can carry a multi-stage task from planning to finished output without a human stepping in to unstick it.

Everyday agentic work: where Sonnet 5 holds its own

Sonnet 5 is built for the work most teams actually run all day. On agentic coding it scores 63.2%, a clear generational jump from Sonnet 4.6 at 58.1%, and it lands within striking distance of Opus 4.8 at 69.2%. On several knowledge-work benchmarks it even edges slightly past Opus 4.8. It is also self-correcting: it reviews its own answer and fixes errors before you ever see them, which matters more than a headline benchmark for production reliability. For teams running thousands of automated agent turns a day, that built-in review loop quietly removes a whole class of silent failures, and it does so without the token bill of a frontier model.

Here is how the current Claude models stack up on agentic coding:

As Anthropic put it when introducing the model:

Fable 5’s capabilities exceed those of any model Anthropic has ever made generally available.

Anthropic

That framing captures the trade-off perfectly: Fable 5 is the new ceiling, but a ceiling you pay a premium to reach.

Pricing: the 5x gap

Price is where the two models diverge most sharply, and it is usually the deciding factor for anyone running these at scale.

Fable 5 pricing

Fable 5 is priced at $10 per million input tokens and $50 per million output tokens. That is flagship pricing for flagship power, and it makes Fable 5 the most expensive model in the current Claude lineup by a wide margin.

Sonnet 5 pricing

Sonnet 5 launched with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it moves to $3 per million input and $15 per million output, per Anthropic’s official pricing documentation. For context, Opus 4.8 sits between them at $5 input and $25 output. The practical upshot is that Fable 5 costs roughly three to five times more per output token than Sonnet 5, so the value question comes down to whether a task genuinely needs the extra ceiling. A rule of thumb: if you can solve a problem on Sonnet 5 without repeated retries, the cheaper model is almost always the economically correct choice, because the price delta compounds fast across large output volumes.

Availability and safety differences

The two models also differ in how they are deployed and guarded. Sonnet 5 shipped as a broadly available default; Fable 5 launched with tighter guardrails and a rockier rollout.

Fable 5: conservative safeguards and a bumpy launch

Fable 5 shipped with a safeguard mechanism: on certain sensitive queries, the response comes from Claude Opus 4.8 instead, and Anthropic tuned this to trigger in fewer than 5% of sessions on average. The launch was also disrupted. On June 12, U.S. export controls forced Anthropic to restrict access, and because nationality could not be verified in real time, the company suspended access for everyone. The controls on Fable 5 were lifted on June 30, and as of July 1, 2026, Fable 5 is available globally again, as Anthropic explained in its redeployment note.

Sonnet 5: default everywhere, strong safety

Sonnet 5 took the opposite path, becoming the default model on Free and Pro plans on release day and shipping in Claude Code and the API. On safety, both new Sonnet models scored 0.0% on the Firefox 147 exploit benchmark, work done with Mozilla and patched in Firefox 148, alongside improved refusal behavior and prompt-injection resistance.

Which one should you choose?

For most people the decision is straightforward once you weigh cost against the ceiling you actually need.

Choose Sonnet 5 for everyday agents and coding at scale

For the majority of workloads, everyday agents, production coding, chat, and budget-conscious automation, Sonnet 5 is the right call. It is three to five times cheaper on output, self-correcting, and already the default on consumer plans. If you just want to try the model without any setup, Sonnet 5 is also available free in the browser through the sonnet5.pro Chrome extension.

Use this quick test to decide:

1. Estimate your monthly output token volume.
2. Ask whether the task is long-horizon and autonomous, or short and repeatable.
3. Check whether accuracy failures are expensive or merely annoying.
4. If the work is routine and volume is high, pick Sonnet 5.
5. If the work is rare, complex, and high-stakes, consider Fable 5.
6. Prototype on Sonnet 5 first, then escalate only the tasks that stall.

Choose Fable 5 for the hardest, longest, highest-stakes tasks

Reach for Fable 5 when maximum capability justifies the price. That means the toughest, longest autonomous runs, deep research, and heavy repository-level engineering where its lead widens with task length. Just budget for the premium output pricing and keep the Opus 4.8 safeguard fallback and the model’s early access restrictions in mind when you build around it.

FAQ