Claude’s Model Lineup in 2026: Which One Should You Actually Use?

Verdict: Anthropic now runs five Claude models: Haiku 4.5 for speed, Sonnet 5 for daily work, Opus 5 for deep reasoning (it quietly replaced Opus 4.8 in July), and Fable 5 for frontier-level tasks, with Mythos 5 locked to vetted research partners. Most people only need Sonnet 5 — save Opus 5 or Fable 5 for problems that actually justify the cost.

---

Claude’s Lineup Right Now

Model	Best for	Context window	Price (input/output per 1M tokens)
Haiku 4.5	High-volume, latency-sensitive tasks	200K	$1 / $5
Sonnet 5	Everyday work — writing, coding, research	1M	$2 / $10 (through Aug 31, then $3 / $15)
Opus 5	Deep reasoning, long agentic runs	1M	$5 / $25
Fable 5	Frontier-tier, hardest reasoning and coding	1M	$10 / $50
Mythos 5	Same model as Fable 5, research-only access	1M	Not publicly priced

Opus 5 replaced Opus 4.8 as Anthropic’s current Opus model on July 24, 2026, at the same $5/$25 rate. If you tested Opus 4.8 a few months ago and are deciding whether to switch, the pricing hasn’t moved — the model behind the name has.

Why the Naming Still Trips People Up

Claude’s names used to track length: Haiku (short), Sonnet (structured), Opus (a major work). That logic broke when Anthropic added a tier above Opus instead of just pushing version numbers higher. Fable and Mythos aren’t “Claude 5” in the way people expect a flagship number to work — they’re a separate, more expensive class sitting on top of the existing lineup, and Fable 5 is the version of it anyone can actually buy. Mythos 5 shares the same underlying model but stays restricted to a small set of organizations through Anthropic’s Project Glasswing.

The Three-Week Gap Most Guides Don’t Mention

Fable 5 and Mythos 5 launched June 9, 2026. Three days later, Anthropic suspended both to comply with U.S. Department of Commerce export controls. The controls lifted June 30, and access came back July 1 — a three-week stretch where the top of the Claude lineup simply wasn’t available, on any plan.

That matters for one practical reason: if a workflow depends on Fable 5 specifically, it has already proven less stable than Sonnet, Opus, or Haiku. Anthropic’s own account of the suspension and restoration is worth reading if you’re deciding how much to lean on it.

Where We’d Actually Point You

We run a deep-research and scriptwriting operation — pulling from dozens of sources per video, then turning that into a script that has to survive a professional client’s review. That’s a different bar than “does this model give a decent answer,” and it’s changed which Claude tier we default to for which stage of the work.

Sonnet 5 is where most of our research and first-draft writing happens now. It’s fast enough that iterating on a script outline doesn’t feel like a wait, and the 1M context window means we can drop an entire batch of source transcripts into one conversation instead of chunking them. We reach for Opus 5 when a topic has enough moving parts that Sonnet starts losing the thread — multi-angle documentary research, for instance, where a dozen different claims need to stay internally consistent across a 3,000-word script. Fable 5 hasn’t become a daily tool for us; the cost only makes sense on the rare piece where the reasoning genuinely has to be frontier-level, and given the June suspension, we don’t build anything time-sensitive around it without a fallback.

(Note to editor: swap this section for your team’s actual test notes and specific examples before this goes live — the framing above is a starting structure, not a substitute for real usage data.)

When Not to Overthink This

• You’re doing everyday work. Sonnet 5 covers most real tasks. Defaulting to the priciest tier out of habit just inflates the bill.
• You need guaranteed uptime. After the June suspension, treat Opus 5 as the fallback for anything mission-critical, not Fable 5 alone.
• You’re chasing the “newest” model on instinct. Fable 5 isn’t next in line after Opus — it’s a separate, pricier tier for specific frontier work, not an automatic upgrade.

Our Testing Method

We cross-checked Anthropic’s own documentation and pricing pages against the current Claude apps and API behavior in August 2026, and reviewed Anthropic’s public statement on the June export-control suspension for accuracy.

Last updated: August 2026