What is the difference between Claude Fable 5, Mythos 5, and Sonnet 5?

Fable 5 is Anthropic's flagship long-horizon coding and agent model, redeployed globally on July 1, 2026 with a new cybersecurity classifier. Mythos 5 is the specialized cybersecurity variant, accessible only to a select group of US organizations after the July 1 export-control adjustment. Sonnet 5 is the newer, cheaper mid-tier model released June 30, 2026 that punches close to Opus 4.8 on agentic coding.

Which Claude model should I use for coding in July 2026?

For most coding work, use Sonnet 5. It scores 63.2% on Anthropic's agentic coding benchmark at $2/$10 introductory pricing (through Aug 31, then $3/$15). Use Opus 4.8 (69.2%) only for the hardest agent runs. Use Fable 5 for very long-horizon multi-hour engineering tasks where its specialized training pays off. Mythos 5 is restricted to authorized cybersecurity users.

Why were Fable 5 and Mythos 5 restricted in June 2026?

In June 2026, Amazon researchers reported a jailbreak in Fable 5. The US Commerce Department briefly imposed export-control restrictions on both Fable 5 and Mythos 5 while Anthropic and the US government negotiated safeguards. Restrictions lifted July 1, 2026 with Anthropic adding a new cybersecurity classifier and additional guardrails.

Can I access Mythos 5 through the Claude API?

No. As of July 2, 2026, Mythos 5 remains restricted to a select group of US organizations vetted through the Project Glasswing program. General API access is limited to Fable 5, Sonnet 5, Opus 4.8, and earlier Claude models. Enterprise customers with cybersecurity needs should contact Anthropic sales.

Quick Answer

Claude Fable 5 vs Mythos 5 vs Sonnet 5 (July 2026)

Published: July 2, 2026

Claude Fable 5 vs Mythos 5 vs Sonnet 5: Which Anthropic Model to Pick (July 2026)

Anthropic had three model events in the last week of June 2026: Fable 5 and Mythos 5 export controls were lifted July 1, and Sonnet 5 launched June 30 as a new mid-tier flagship. That’s three Claude models with very different access rules, pricing, and best-fit use cases. Here’s how to pick the right one for what you’re building.

Last verified: July 2, 2026

At a glance

Model	Positioning	Availability	Pricing	Released
Claude Sonnet 5	Mid-tier flagship for agentic coding	Fully GA (all plans, API, Bedrock, Copilot)	$2/$10 intro, $3/$15 from Sep 1	June 30, 2026
Claude Fable 5	Long-horizon engineering flagship	Globally redeployed July 1, 2026	Premium tier	Earlier 2026
Claude Mythos 5	Cybersecurity specialist	Restricted to authorized US orgs	Enterprise contract	Earlier 2026
Claude Opus 4.8	Highest-capability general model	Fully GA	Opus-tier pricing	Earlier 2026

Claude Sonnet 5 — the everyday workhorse

Anthropic released Sonnet 5 on June 30, 2026 as “the most agentic Sonnet model yet.” The pitch: near-Opus intelligence at Sonnet pricing.

63.2% on Anthropic’s agentic coding benchmark (vs. Opus 4.8 at 69.2%)
Multimodal input, native tool use for browsers and terminals
$2 input / $10 output per 1M tokens introductory (through Aug 31, 2026)
$3 / $15 from September 1
Available on Claude Free, Pro, Max, Team, Enterprise, the API, AWS Bedrock, and GitHub Copilot

Best for: Almost all coding work in July 2026. It’s the drop-in replacement for Sonnet 4.6, and unless you’re on the hardest agent runs, you don’t need Opus.

Claude Fable 5 — the long-horizon engineering flagship

Fable 5 is Anthropic’s flagship for very long engineering runs — the model Stripe used to compress “months of engineering into days” on a 50-million-line Ruby codebase migration.

Redeployed globally on July 1, 2026 after the US export-control adjustment
New cybersecurity classifier added as part of the redeployment safeguards
Sits above Sonnet 5 for the very hardest multi-hour engineering tasks

Best for: Multi-hour autonomous coding sessions, large codebase migrations, and any task where you’d previously reach for Opus but want Fable’s specialized long-horizon behavior.

Claude Mythos 5 — the restricted cybersecurity model

Mythos 5 is Anthropic’s specialized cybersecurity variant, positioned for defensive security work. It’s the flagship of Project Glasswing — Anthropic’s collaborative program with AWS, Google, Microsoft, Nvidia, Cisco, CrowdStrike, Palo Alto Networks, and others.

Access restricted to a select group of US organizations as of July 1, 2026
Redeployed with new safeguards after Amazon researchers reported a jailbreak in June 2026
Not accessible via the general Claude API

Best for: Vulnerability research, incident response, and offensive-security-adjacent work — for organizations pre-approved through Project Glasswing.

Claude Opus 4.8 — the highest-capability generalist

Still Anthropic’s most capable general-purpose model.

69.2% on Anthropic’s agentic coding benchmark
Best for the hardest reasoning tasks
Opus-tier pricing (significantly above Sonnet 5)

Best for: Cases where Sonnet 5 falls short. In July 2026 those cases are narrower than they were three months ago.

Decision guide

Use Sonnet 5 if:

You want the best price/performance for coding, chat, or agent work
You were on Sonnet 4.6 — this is a drop-in upgrade
You care about the $2/$10 introductory pricing window (through Aug 31)

Use Opus 4.8 if:

Sonnet 5 is failing on your hardest tasks
You have budget for premium pricing on genuinely hard work

Use Fable 5 if:

You’re running multi-hour autonomous coding sessions
You have a large-codebase migration or transformation project
Sonnet 5 and Opus 4.8 both struggle with the long-horizon coherence

Use Mythos 5 if:

You’re an authorized Project Glasswing member
Your team does defensive cybersecurity work
Otherwise, you can’t use it

What changed on July 1, 2026

The US Commerce Department lifted export-control restrictions on Fable 5 and Mythos 5 that were imposed in June after an Amazon-reported jailbreak. Anthropic added:

A new cybersecurity classifier to Fable 5 as an additional guardrail
Additional safeguards to Mythos 5 access
A commitment to work with Project Glasswing partners on a consensus framework for evaluating AI jailbreak severity

Fable 5 is now globally available. Mythos 5 remains selectively restricted.

Bottom line

For 99% of developers, the answer is Sonnet 5 — cheapest, best price/performance, fully available, and near-Opus on agentic coding. Reach for Opus 4.8 only when Sonnet 5 falls short. Use Fable 5 for long-horizon engineering runs. Mythos 5 is not an option unless you’re already in Project Glasswing. If in doubt, start with Sonnet 5 and only escalate when a task genuinely demands more.