Claude 5 Family: Fable 5, Sonnet 5 and Opus 5 Explained

The short version#

Anthropic released three Claude 5 models between June 9 and July 24, 2026, and they are not three sizes of the same thing. Fable 5 ($10/$50 per million tokens) is the frontier model with safety classifiers bolted on. Sonnet 5 ($2/$10 until August 31, then $3/$15) is the new default for everyone and the price-performance pick. Opus 5 ($5/$25) sits in between and is the one most working teams should actually evaluate first: Anthropic's own numbers put it near Fable 5 on coding at half the price.

There is a fourth name floating around, Mythos 5, which is not a separate model. It is Fable 5 without certain safety classifiers, available only to vetted partners. If you are choosing a model for your business, Mythos is not on your menu, so this post covers the three you can buy.

Timeline: seven unusually eventful weeks#

The Claude 5 rollout was messier than a normal model launch, and knowing the timeline helps decode confusing coverage from June:

Fable 5: the frontier, with a chaperone#

Fable 5 is what Anthropic calls a "Mythos-class" model made safe for general availability. Per the announcement, it outperforms every previously available Claude model on software engineering, knowledge work, vision and scientific research. The launch material leads with agentic endurance: Stripe reportedly compressed a migration estimated at two months into a single day, and the model completed Pokémon FireRed unassisted in a vision-based harness.

The distinctive part is the safety architecture. Fable 5 runs new classifiers that detect potential misuse in cybersecurity, biology/chemistry and model distillation. When they trigger, the request falls back to Claude Opus 4.8. Anthropic says this affects fewer than 5% of sessions, but it is a real behavioral difference: the same prompt can be answered by a different model depending on topic.

Pricing: $10 input / $50 output per million tokens, which Anthropic notes is less than half the old Mythos Preview price. That is still 5x Sonnet 5's intro input price.

Mythos 5 is the same underlying model with some of those classifiers relaxed, distributed through Project Glasswing (a US government cybersecurity cooperation) and a planned Trusted Access Program for biology researchers. Anthropic describes it as having the strongest cybersecurity capabilities of any model in the world. Unless you are one of those partners, you interact with this capability tier only through Fable 5.

Sonnet 5: the new default, priced to move#

Sonnet 5 replaced Sonnet 4.6 as the default for every Free and Pro Claude user on June 30. Anthropic pitches it as its "most agentic Sonnet yet," and MarkTechPost's comparison collects the published numbers: 63.2% on SWE-bench Pro, 81.2% on OSWorld-Verified and 57.4% on HLE, close enough to Opus 4.8 that the gap stops mattering for most day-to-day work. On Anthropic's knowledge-work benchmark, Sonnet 5 actually edges out Opus 4.8.

The pricing is openly promotional: $2 input / $10 output until August 31, 2026, then $3/$15. The New Stack called it what it is, a land-grab price during the GPT-5.6 launch window. If you are running high-volume agent workloads, the arithmetic favors migrating before the September price step, not after.

Opus 5: the one to actually evaluate#

Opus 5 arrived on July 24 at the same $5/$25 price as Opus 4.8, and the headline from Anthropic's announcement is aggressive: near-Fable performance at half the price. The published claims worth knowing:

• Frontier-Bench v0.1: tops all models and more than doubles Opus 4.8's score at a lower cost per task.
• CursorBench 3.2 (max effort): 0.5 points below Fable 5 at half the cost.
• ARC-AGI 3: three times the next-best model's score.
• OSWorld 2.0 (computer use): beats Fable 5 at roughly a third of the cost.
• Zapier AutomationBench: ~1.5x the success rate of the next-best model.

Two new API features shipped in beta alongside it: mid-conversation tool changes and automatic fallbacks. There is also a Fast Mode at roughly 2.5x speed for double the base price.

All of these are vendor-published numbers, not independent replications. But the pattern across them is consistent: Opus 5 is positioned as the rational default for demanding work, with Fable 5 reserved for the cases where the last few percent matter.

Which one for which job#

The uncomfortable question for Anthropic's own lineup: with Opus 5 this close to Fable 5 on published coding and computer-use numbers, the $10/$50 tier is a hard sell for anything except research-grade workloads. That is presumably why subscription plans meter Fable 5 usage separately (credits beyond a weekly allowance, per Forbes).

What this post does not cover#

We have not independently benchmarked any of these models; every performance number above is from Anthropic's own announcements or the cited third-party write-ups. Context-window sizes for the Claude 5 family were not stated in the launch posts we reviewed, so we omit them rather than guess. And Mythos 5's actual capabilities are only knowable to Glasswing partners.

Sources#

• Anthropic: Claude Fable 5 and Claude Mythos 5
• Anthropic: Introducing Claude Sonnet 5
• Anthropic: Introducing Claude Opus 5
• Anthropic: Redeploying Claude Fable 5
• TechCrunch: Anthropic launches Claude Sonnet 5
• The New Stack: Sonnet 5 launch analysis
• MarkTechPost: Sonnet 5 vs Sonnet 4.6 vs Opus 4.8
• MacRumors: Fable 5 available again
• Forbes: Anthropic extends Fable access