Cafiyn Pulse
Anthropic · Verified 2026-09-05

Claude Opus 5 pricing, and when the premium is worth it.

Anthropic's top reasoning tier costs 2.5x Sonnet 5. That is a much smaller gap than the previous generation, which changes the routing maths.

Claude Opus 5
Anthropic · premium tier
Verify against Anthropic's pricing page ↗
$5.00
per 1M input tokens
$25.00
per 1M output tokens
A typical 1,200-in / 400-out token request costs about $0.0160, or roughly $16.00 per 1,000 requests.

Opus 5 runs $5 per million input tokens and $25 per million output, which is two and a half times Sonnet 5 on both legs. The number worth noticing is what it replaced: Opus 4.8 was $15 and $75, so the current top tier costs a third of the outgoing one. If your routing config still names Opus 4.8, you are paying three times over for the older model, and that is a one-line change.

That price drop changes the calculus. Under the old 5x multiplier, escalating to Opus was a decision you made carefully for a small slice of traffic. At 2.5x it is defensible for a much wider set of calls, particularly multi-step agent work where one wrong intermediate step compounds into a wasted run and you pay for the retry anyway. Model the retry rate, not just the per-token price.

Best for
  • Complex multi-step agentic tasks where a wrong intermediate step compounds
  • High-stakes generation where the cost of an error outweighs the API cost
  • The escalation tier in a routed system, though now a cheaper escalation than it used to be
Head to head

Claude Opus 5 vs the alternatives.

ModelInput / 1MOutput / 1MSample request
Claude Opus 5$5.00$25.00$0.0160
GPT-6 Astra$10.00$50.00$0.0320
Claude Sonnet 5$2.00$10.00$0.0064
Sample request assumes 1,200 input and 400 output tokens, a typical chat exchange. Your actual ratio will differ, run your own numbers in the calculator below.
FAQ

Common questions.

How much does Claude Opus 5 cost?

$5 per million input tokens and $25 per million output tokens. That is 2.5x Claude Sonnet 5 on both legs, and a third of the price of the outgoing Opus 4.8, which Anthropic has moved to its legacy list at $15 and $75.

Is Claude Opus 5 worth the extra cost over Sonnet 5?

More often than the previous generation was. At 2.5x rather than 5x, Opus earns its price on tasks where reasoning depth measurably changes the outcome: complex multi-step agentic work, or generation where an error is expensive to correct. For routine production traffic Sonnet 5 still handles it fine at 40 percent of the cost.

Should I still be using Claude Opus 4.8?

Almost certainly not. Anthropic lists Opus 4.8 as legacy at $15 input and $75 output, three times the price of Opus 5. Unless you have pinned to it for a reproducibility reason, switching is a straight cost reduction.

More models

Browse other pricing pages.

Open the tool.

Live math against Anthropic's current pricing plus every other provider in the roster.

Compare Claude Opus 5 on my usage