Claude Opus 5 Is Here: What Professionals Need to Know (July 2026)
Anthropic launched Claude Opus 5 on July 24, 2026 — near Fable 5's frontier intelligence at half the price. Here's what changed, what the benchmarks actually mean, how the new effort toggle works, and whether you need to do anything.
See Claude set up for your job
Skip the theory — pick your profession and get the real workflows, ready-to-use prompts, and exact setup for your work.
Update (September 22, 2026): Claude Opus 5.5 launched today — list price $4/$20 per million tokens (20% below Opus 5), about 40% cheaper to run on typical workloads, matching Fable 5.1's performance on most work. Opus 5.5 is now the current Opus on Pro, Max, Team, and Enterprise. If you use claude.ai, you have it automatically. API users: update your model string to
claude-opus-5-5. Details: Claude Opus 5.5 for professionals.
Update (September 1, 2026): Claude Fable 5.1 shipped on September 1, 2026 and now sits above Opus 5 as Anthropic's most capable generally available model (same $10/$50 as Fable 5, cache reads cut to $0.25). Details: Claude Fable 5.1 for professionals.
TL;DR. Anthropic launched Claude Opus 5 on July 24, 2026. It delivers close to Fable 5's frontier capability at half Fable 5's API price — and on several benchmarks it actually exceeds Fable 5. It is available now on Pro, Max, Team, Enterprise, Claude Code, Claude Cowork, and the API (
claude-opus-5). API pricing is unchanged from Opus 4.8: $5/$25 per million tokens.
Anthropic launched Claude Opus 5 today. It is the fifth generation of the Opus line — the company's deepest-reasoning tier — and the first to close the gap with Fable 5 on real-world work benchmarks, at half Fable 5's token price.
Here is what changed, what the benchmarks mean in practice, and what to do.
Where Opus 5 fits in the lineup
Claude's model lineup now has four generally available tiers:
| Model | Position | API pricing |
|---|---|---|
| Fable 5 | Frontier (Max/Team Premium, credits) | $10/$50 per M tokens |
| Opus 5 | Most capable on Pro; default on Max | $5/$25 per M tokens |
| Sonnet 5 | Default (Free, Pro) | $2/$10 per M tokens |
| Haiku 4.5 | Fast / high-volume | $1/$5 per M tokens |
Opus 5 positions itself squarely between Sonnet 5 and Fable 5 in cost — and in many respects in capability too, while exceeding Fable 5 on the benchmarks most relevant to knowledge work. Haiku 5 is the one remaining Series 5 release still outstanding.
What actually changed
The headline claim from Anthropic is that Opus 5 is "a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price." That's the marketing line. Here is what the data show.
Self-verification is dramatically better. Anthropic's own demos show Opus 5 building a computer vision pipeline from an incomplete prompt — generating the tool it needed when it didn't have one, then checking its own output and iterating. The model is described as "much stronger at verifying its work and iterating carefully until it succeeds," which is the practical behavior that separates a useful agent from one that stalls or hallucinates mid-task.
Fewer friction points. Safety filters on Opus 5 trigger approximately 85% less frequently than on Fable 5, and when they do trigger, a new beta feature automatically routes the request to a less capable model rather than returning a hard error. In practice: fewer refusals on legitimate professional work, and less session breakage when a classifier fires.
Effort control. Opus 5 ships with the same effort settings as Sonnet 5 — low, medium, or high — letting you optimize for cost and speed on lighter queries while reserving full reasoning depth for complex tasks. This is the "toggle between cost and capability" feature that Anthropic highlighted at launch.
The benchmarks
These come from Anthropic's published data and launch reporting, verified across multiple outlets:
| Benchmark | Opus 5 | Opus 4.8 | Fable 5 |
|---|---|---|---|
| Frontier-Bench v0.1 (agentic terminal coding) | 43.3% | ~21% | 33.7% |
| GDPval-AA v2 (knowledge work) | 1,861 Elo | — | 1,747 |
| ARC-AGI-3 (novel problem-solving) | 30.2% | — | — |
What these say, translated:
Agentic coding: Opus 5 at 43.3% is approximately double Opus 4.8's performance and beats Fable 5's 33.7%. For professionals using Claude in agentic or coding workflows, this is the headline number — a big jump from what Opus 4.8 could do, and ahead of even the flagship model.
Knowledge work: The 1,861 Elo on GDPval-AA v2 surpasses Fable 5's 1,747. For lawyers drafting, consultants analyzing, financial professionals summarizing — everyday document-heavy professional tasks — Opus 5 benchmarks above Fable 5 at half the price.
Novel problem-solving: ARC-AGI-3 tests reasoning on genuinely unfamiliar problems. Opus 5's 30.2% is roughly four times higher than GPT-5.6 Sol on the same benchmark, according to Anthropic's reporting.
Where Opus 5 is not ahead: Fable 5 retains an edge on some domains — particularly the longest, most complex multi-day autonomous projects, and certain health and legal specialized tasks. If you're running Fable 5 for weeks-long agentic workflows, the recommendation still stands to stay there. On the specific benchmarks for cybersecurity exploitation, Opus 5 also falls behind Fable 5 (and Anthropic notes stronger guardrails in this area).
Pricing
API:
- Standard: $5 per million input tokens, $25 per million output tokens — unchanged from Opus 4.8
- Fast mode: approximately 2× the base price for 2.5× the response speed
- Compare to Fable 5: $10/$50 per million tokens — Opus 5 is exactly half
Claude.ai:
- Included in existing Pro, Max, Team, and Enterprise subscriptions
- On Claude Max: Opus 5 is now the default model (replacing Fable 5 as the subscription default)
- On Claude Pro: Opus 5 is the most capable model available (Fable 5 on Pro moved to credits-based billing in July; Opus 5 is now the practical ceiling for the $20/month plan)
- Free plan: Opus 5 is not included
The economics here matter for API developers in particular. Opus 5 delivers meaningfully more than Opus 4.8 at the same per-token price — effectively a free capability upgrade if you update your model string.
Who benefits most
API developers using Opus 4.8. The clearest win: swap claude-opus-4-8 for claude-opus-5, same price, substantially better performance on agentic and knowledge-work tasks.
Claude Code and Claude Cowork users on Max. Opus 5's stronger agentic capability and self-verification make it better suited to multi-step coding and workflow tasks. Claude Cowork runs on Max by default — Opus 5 is available there now.
Pro subscribers who have been paying Fable 5 credits for everyday work. With Opus 5 benchmarking above Fable 5 on knowledge work, the case for burning Fable 5 credits on document drafting and analysis is weaker. Test Opus 5 first.
Long-running autonomous projects. If you've been running Fable 5 specifically because Opus 4.8 stalled too often on extended tasks, Opus 5's self-verification improvements are worth testing. It may close enough of the gap that Fable 5 credits are only needed for the genuinely complex multi-day runs.
What to do
Claude.ai users (Pro, Max, Team, Enterprise): Nothing required. Opus 5 is your Opus-tier model as of today. On Pro, it is accessible via the model picker. On Max, it is the new default.
API users with pinned model strings: Update from claude-opus-4-8 to claude-opus-5. Existing pinned strings will continue to route for now, but migrate before Anthropic announces the Opus 4.8 deprecation timeline.
Claude Code users: Update to claude-opus-5 if you had Opus 4.8 or 4.7 pinned. The agentic improvements make this the better choice for multi-file and long-context coding tasks.
Sources
- Anthropic official announcement: Claude Opus 5 (July 24, 2026)
- The Decoder: Anthropic claims its new Claude Opus 5 delivers near Fable 5 performance at half the token price (July 24, 2026)
- TechCrunch: Anthropic launches Opus 5 (July 24, 2026)
- Fortune: Anthropic debuts Claude Opus 5 with feature that lets users toggle between cost and capability (July 24, 2026)
See Claude set up for your job
Skip the theory — pick your profession and get the real workflows, ready-to-use prompts, and exact setup for your work.
Set up AI for your job — free, in about 2 minutes
Pick your profession and get your first working AI tool, a step-by-step guide, and a $0 plugin to take home. No credit card.
Get my free setupSee Claude set up for your job
Real workflows and ready-to-use prompts, profession by profession.
Frequently asked questions
What is Claude Opus 5?+
Claude Opus 5 is Anthropic's newest Opus-tier model, launched July 24, 2026. It replaces Opus 4.8 as the most capable model on Claude Pro and becomes the default on Claude Max. Anthropic describes it as delivering near Fable 5's frontier intelligence at half Fable 5's token price. The API model ID is `claude-opus-5`.
How much does Claude Opus 5 cost?+
API pricing: $5 per million input tokens and $25 per million output tokens — the same as Opus 4.8, and half the price of Fable 5 ($10/$50). Fast mode is available at approximately 2× the base price for 2.5× the response speed. On claude.ai, Opus 5 is included in your existing Pro, Max, Team, or Enterprise subscription — no additional charge.
Is Claude Opus 5 better than Fable 5?+
On several benchmarks, yes — Opus 5 actually surpasses Fable 5. It scores higher on Frontier-Bench v0.1 agentic terminal coding (43.3% vs Fable 5's 33.7%), GDPval-AA v2 knowledge work (1,861 Elo vs Fable 5's 1,747), and ARC-AGI-3 novel problem-solving (30.2%, roughly 4× higher than GPT-5.6 Sol). It falls behind Fable 5 on some tasks — particularly long multi-day autonomous projects and certain health and legal domains. Anthropic still recommends Fable 5 for the most complex extended autonomous work.
What is the effort toggle in Claude Opus 5?+
Opus 5 includes user-controlled effort settings (low, medium, or high) that let you trade reasoning depth for speed and cost. Set effort high for complex, multi-step tasks where quality matters; drop to low or medium for simple queries where you want a faster, cheaper response. This is the same effort-control system already available on Sonnet 5.
Do I need to do anything when Claude Opus 5 launches?+
Claude.ai users (Pro, Max, Team, Enterprise): nothing — Opus 5 becomes your Opus-tier model automatically. Claude Code and Claude Cowork users on Max: Opus 5 is available now. API users who pinned `claude-opus-4-8` or `claude-opus-4-7`: update your model string to `claude-opus-5` to get the new model. Free plan users: Opus 5 is not on the Free plan.
What happened to Opus 4.8?+
Opus 4.8 is succeeded by Opus 5. Existing API strings pointing to `claude-opus-4-8` will continue to work for now, but Anthropic will announce a deprecation timeline. On claude.ai, Opus 5 replaces Opus 4.8 as the Opus-tier option. Haiku 4.5 remains the Haiku-tier model — Haiku 5 is the only Series 5 model still unannounced.
Related Guides
What Is Claude Duet, and Should You Join the Waitlist? (October 2026)
Anthropic opened a research preview of Claude Duet on October 21, 2026: a separate app where Claude works as a proactive agent, sending a morning brief, watching your connected apps, and asking before it sends, pays, signs, or deletes anything. Here's what it does, who can get it, what it costs, and how to try it safely.
How Do I Lock a Chat in ChatGPT? What Locked Chats Hide — and What They Don't
OpenAI began rolling out Locked Chats on October 20, 2026. You can put individual ChatGPT conversations behind a PIN or Face ID/fingerprint so they drop out of your history and search. Here's how to turn it on, what it protects, and the PIN-reset catch that permanently deletes everything.
Should You Let an AI Agent Into Your Bank Account? Why Chase, Citi, BofA and Schwab Are Saying No (October 2026)
In mid-October 2026, JPMorgan Chase, Citi, Bank of America and Charles Schwab all said outside AI agents like Meta's Muse cannot log in to customer accounts or move money with a customer's password. Here's what each one blocks, what still works (read-only connections), who is liable if an agent gets it wrong, and how to decide what an AI agent should be allowed to touch.