Claude Opus 5: What It Is, What It Costs, and Where to Use It (2026)

Quick Answer (updated August 16, 2026): Anthropic released Claude Opus 5 on July 24, 2026 as model ID claude-opus-5. It is priced at $5 per million input tokens and $25 per million output tokens — identical to Opus 4.8 — and Anthropic says it outperforms Fable 5 on a number of benchmarks despite Fable 5 sitting a tier above it. That combination is the story: this is the first Anthropic release in 2026 to raise the practical capability ceiling without raising the price. Use Opus 5 as your default for agentic and everyday work; reserve Fable 5 for specialist research ceilings, and Sonnet 5 for cheap mechanical edits.

What Is Claude Opus 5?

Claude Opus 5 is Anthropic's Opus-tier model released on July 24, 2026, succeeding Opus 4.8 (May 28, 2026). It sits below the Mythos-class Fable 5 in Anthropic's lineup on paper — and above it on several practical benchmarks.

The headline detail is the pricing. Opus 5 costs $5 per million input tokens and $25 per million output tokens, exactly what Opus 4.8 cost. Anthropic did not charge for the upgrade.

Anthropic's own framing at launch was that Opus 5 "outperforms Fable 5 on a number of benchmarks," which is unusual — a lower tier beating the flagship. The explanation is that Opus 5 was tuned for a different thing: verifying its own work and iterating until it succeeds, rather than reaching a higher single-shot ceiling. Anthropic's launch example was the model writing a complete computer vision pipeline from an incomplete prompt.

Claude Opus 5 at a Glance

Two of those rows matter more than the benchmark table for anyone deploying this in production:

Automatic Fallbacks. When a safety system triggers, the request is routed to a less powerful model instead of failing outright. Fable 5's equivalent behaviour was a hard downgrade to Opus 4.8 on high-risk classification, which broke workflows silently. Fallbacks make the degradation graceful.

The retention difference. Fable 5 carries a 30-day data retention policy; Opus 5 does not. For teams with contractual data-handling constraints, that single line is often the whole decision, independent of benchmarks.

The 2026 Anthropic Timeline, in Order

Opus 5 makes more sense against the year that produced it:

The Fable 5 suspension is the context most coverage skips. Between June 12 and June 30, the frontier tier was unavailable to everyone, and teams that had built on it fell back to Opus 4.8. Opus 5 landed three weeks later at Opus pricing with most of the capability — which is, in practice, the answer to the question that suspension raised: how much do you actually lose by not being on the top tier?

Less than you would think.

Where Claude Opus 5 Is the Right Choice

1. Agentic and Long-Running Work

This is Opus 5's designed strength. It leads on terminal coding (reported at 43.3% on Frontier-Bench), novel problem-solving (30.2% on ARC-AGI-3), and business-workflow automation. The "verify and iterate" tuning is what makes it hold up across long autonomous runs where a single wrong turn used to compound.

2. Everyday High-Quality Work at Predictable Cost

At the same price as its predecessor with a higher ceiling, Opus 5 is the straightforward default for document analysis, drafting, research synthesis, and coding. There is no cost argument for staying on Opus 4.8.

3. Work With Data-Handling Constraints

Because it sits outside the 30-day retention policy that applies to Fable 5, Opus 5 is the practical frontier-adjacent option for regulated or contractually constrained workloads.

4. Anything Cyber-Defensive

Opus 5's safeguards restrict exploitation-oriented output while preserving vulnerability identification — the intended users are defenders and product security teams. Fable 5's broader classifiers made this class of work frustrating; Opus 5's engage an estimated 85% less often.

Where Opus 5 Is Not the Right Choice

• Specialist research ceilings. Fable 5 remains stronger on domain benchmarks such as legal and health. If you are chasing the maximum on a narrow specialist task, the flagship still wins.
• Cheap, mechanical, high-volume edits. Sonnet 5 is the economical answer, and using Opus 5 for classification or short summarization wastes money on reasoning you did not need.

Opus 5 vs Fable 5 vs Sonnet 5

The honest summary: Opus 5 is the one to default to, and Fable 5 is now a specialist tool rather than a general upgrade. That is a genuine reversal from June, when the Mythos tier was framed as strictly better.

What This Means for Making Presentations

Model choice matters less for slides than most people assume, and it matters in a specific, narrow way.

Layout, typography, chart rendering, and export quality are properties of the tool, not the model — a stronger model will not make a deck look better. What a stronger model changes is the reasoning inside the content: whether the argument survives a hostile question, whether contradictory sources get reconciled instead of averaged, and whether the numbers in a chart actually follow from the source document.

That is why picking a model should be a per-deck decision rather than an account setting. A class talk and a board-level market analysis do not need the same reasoning depth, and paying for the deepest model on both is simply waste.

ChatSlide already supports Anthropic's Claude Opus, alongside the GPT-5.6 family (Luna, Terra, and Sol), and lets you choose the model per deck rather than locking you to whatever the vendor picked. You write a real brief — what the presentation should accomplish, who is in the room, what they should decide — and the model works against that, not against a one-line topic string. Every deck exports to PowerPoint, Keynote, or PDF.

If you want to see the difference model choice actually makes on slide output rather than on a coding benchmark, our benchmark of the best AI model for presentations tests them on decks instead.

The Bottom Line

Claude Opus 5 shipped July 24, 2026 at $5/$25 per million tokens — the same price as Opus 4.8 — and beats Fable 5 on several benchmarks while sitting a tier below it. It carries no 30-day data retention policy, triggers safety classifiers an estimated 85% less often than Fable 5, and adds Automatic Fallbacks so blocked requests degrade instead of failing.

For most teams, Opus 5 is now the correct default, Fable 5 is a specialist instrument, and Sonnet 5 is the volume workhorse.

Try ChatSlide free — build a presentation with a top-tier model in under 60 seconds, and switch models per deck rather than per account.

---

Model details reflect Anthropic's July 2026 announcements and contemporaneous reporting: the July 24, 2026 launch of Claude Opus 5 (TechCrunch, Axios), the 2026 Claude release timeline (Wikipedia), and reported pricing and benchmark figures. Benchmark numbers are as reported and have not been independently reproduced by ChatSlide. Last updated August 16, 2026.