Which is cheaper, Claude Opus 4.8 or Claude Sonnet 5?

Claude Sonnet 5. On a task of 1,500 input and 500 output tokens across 3 calls, Claude Sonnet 5 costs $0.02400 per task against $0.06000 for Claude Opus 4.8, which makes it 2.50x cheaper (a 60% saving).

How much do Claude Opus 4.8 and Claude Sonnet 5 charge per million tokens?

Claude Opus 4.8: $5 per million input tokens and $25 per million output tokens. Claude Sonnet 5: $2 input and $10 output. These are published list rates as of August 2026, excluding cached, batch and long-context tiers.

What does a million tasks cost on Claude Opus 4.8 vs Claude Sonnet 5?

At the same fixed profile, one million tasks cost $60,000 on Claude Opus 4.8 and $24,000 on Claude Sonnet 5. The gap comes entirely from list-price differences, not from any difference in the work done.

Model price comparisons · August 2026

Claude Opus 4.8 vs Claude Sonnet 5: API pricing compared

Same task, same tokens, two invoices. Every figure on this page is computed from our open pricing dataset, not copied from a blog post.

Claude Sonnet 5 is 2.50x cheaper than Claude Opus 4.8 on a realistic task profile of 1,500 input and 500 output tokens per call, 3 calls per task: $0.02400 against $0.06000 per task, a 60% saving on identical work. Run a million tasks a month and that is the difference between $24,000 and $60,000.

Model	Input $/1M tok	Output $/1M tok	Output multiple	$/task	$/1M tasks
Claude Opus 4.8Anthropic · frontier	$5	$25	5x	$0.06000	$60,000
Claude Sonnet 5Anthropic · mid	$2	$10	5x	$0.02400	$24,000

Monthly bill at three volumes

Tasks / month	Claude Opus 4.8	Claude Sonnet 5	Gap
10,000	$600	$240	$360
100,000	$6,000	$2,400	$3,600
1,000,000	$60,000	$24,000	$36,000

Where the money goes on each

Claude Opus 4.8 prices output at 5x its input rate, so on this profile output is 63% of its task cost. Claude Sonnet 5 prices output at 5x input, putting output at 63% of cost. The model with the higher output multiple rewards capping response length more; the one with the higher input share rewards prompt caching and context trimming more. Cost tells you nothing about quality: run both on twenty of your real cases before deciding, and treat the cheaper model as the default until it demonstrably fails.

Break-even against a dedicated GPU

Against a $2/hour dedicated instance ($1,460/month fixed), self-hosting overtakes Claude Opus 4.8 at 24,333 tasks per month and overtakes Claude Sonnet 5 at 60,833 tasks per month. The full method is in the self-hosting break-even guide.

Pricing the self-hosted side? Compare on-demand GPU rates:

Compare GPU cloud pricing →

Plug your own token counts into the $/task calculator, or read the full ten-model pricing comparison.

Compare against other models

Frequently asked

Which is cheaper, Claude Opus 4.8 or Claude Sonnet 5?: Claude Sonnet 5. On a task of 1,500 input and 500 output tokens across 3 calls, Claude Sonnet 5 costs $0.02400 per task against $0.06000 for Claude Opus 4.8, which makes it 2.50x cheaper (a 60% saving).
How much do Claude Opus 4.8 and Claude Sonnet 5 charge per million tokens?: Claude Opus 4.8: $5 per million input tokens and $25 per million output tokens. Claude Sonnet 5: $2 input and $10 output. These are published list rates as of August 2026, excluding cached, batch and long-context tiers.
What does a million tasks cost on Claude Opus 4.8 vs Claude Sonnet 5?: At the same fixed profile, one million tasks cost $60,000 on Claude Opus 4.8 and $24,000 on Claude Sonnet 5. The gap comes entirely from list-price differences, not from any difference in the work done.