Anthropic shipped Claude Opus 5 on July 24, 2026 — the fourth release in the Claude 5 family in under two months, following Sonnet 5's debut on June 30. It becomes the new default model for Claude Max, and it does so without raising the price tag: Opus 5 launches at the same $5/$25 per million token rate (input/output) as Opus 4.8.
The headline pitch is a new state-of-the-art on Frontier-Bench and GDPval-AA — Anthropic's benchmarks for coding and knowledge work — plus a feature that changes how you pay for reasoning: the effort dial.
What Changed With Opus 5
- New SOTA on coding and knowledge benchmarks — Opus 5 sets new highs on Frontier-Bench and GDPval-AA, Anthropic's internal measures for coding and knowledge-work tasks.
- The effort dial — a new cost/capability toggle that lets you trade reasoning depth for spend on a per-request basis, instead of picking a fixed model tier and living with it.
- Default for Claude Max — new conversations on Claude Max now route to Opus 5 automatically. Free, Pro, and API access continue to default to Sonnet 5 unless you explicitly select Opus.
- No price increase — Opus 5 holds Opus 4.8's $5/$25 per million token pricing rather than launching at a premium, breaking the pattern of each new flagship costing more than the last.
Claude Opus 5 Pricing
| Model | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Claude Opus 5 | $5 | $25 |
| Claude Opus 4.8 | $5 | $25 |
| Claude Sonnet 5 (thru Aug 31, 2026) | $2 | $10 |
If you're already budgeting for Opus 4.8 usage, nothing changes on the invoice side after an Opus 5 migration — same per-token rate, more capability per call.
The Effort Dial: What It Actually Does
Opus 5's effort dial is a cost/capability toggle at the request level rather than the model level. Instead of choosing "Opus for hard tasks, Sonnet for everything else" and living with that split across your whole codebase, you adjust how much reasoning effort a given call spends — dialing up for a gnarly architectural decision, dialing down for routine work — without switching models.
For teams running Opus in production, this is the more consequential change than the benchmark numbers: it turns model selection from an architecture decision into a per-call parameter.
Opus 5 vs Sonnet 5 vs Opus 4.8
- Opus 5 — new default for Claude Max. Best pick for coding and knowledge-work tasks where you want the current ceiling on capability, with the effort dial giving you a way to control cost per request instead of committing to Opus pricing for every call.
- Sonnet 5 — still the default for Free, Pro, and most API workloads. Near-Opus agentic performance at a third of the price; see our Sonnet 5 guide for the full breakdown.
- Opus 4.8 — no reason to start new work here now that Opus 5 matches its price and beats its benchmarks. Existing integrations don't need an emergency migration, but new projects should default to Opus 5.
For the deeper "when do I actually need Opus tier" decision framework, see our Sonnet vs Opus guide and our Opus 4.8 review — most of that logic still applies with Opus 5 in the top slot.
Should You Upgrade?
- On Opus 4.8 already: low-risk swap. Same pricing, better benchmarks, plus the effort dial for cost control. Test against your eval suite before touching production, same as any model swap.
- On Claude Max: you're already on Opus 5 by default — no action needed.
- On Sonnet 5 for cost reasons: stay put unless you're hitting a capability ceiling. The effort dial doesn't change Sonnet 5's price advantage for high-volume agentic work.
The Bottom Line
Opus 5 is Anthropic's fourth Claude 5-family release in under two months, and the pattern holding across all of them is the same: newer, more capable, and priced at or below the model it replaces. The effort dial is the part worth paying attention to beyond the benchmark headline — it's the first sign that "which model tier" is becoming a dial instead of a decision you make once and live with.