Anthropic lists Opus 5 at $5 per million input tokens and $25 per million output tokens, the same listed base API rate as Opus 4.8. Anthropic’s documentation says thinking is enabled by default, which can change token budgets, output limits, and production behavior. Teams should test real workflows before switching.
Anthropic Opus 5 has a useful practical hook for API teams: Anthropic lists the same base API rate as Opus 4.8.
The listed rate is $5 per million input tokens and $25 per million output tokens. Anthropic also lists a Fast mode at twice those base rates: $10 per million input tokens and $50 per million output tokens.
That makes Opus 5 worth testing. A newer model that improves a team’s real work without raising the listed base rate could be a useful upgrade.
It is not a reason to change production systems blindly.
Anthropic’s documentation says thinking is enabled by default in Opus 5. That can affect how requests behave, including token use, response time, visible output, and whether a request reaches its configured output limit.
What teams should notice
The implementation details matter more than a benchmark screenshot:
- API model ID: `claude-opus-5`.
- Listed base pricing: $5 per million input tokens and $25 per million output tokens.
- Thinking is enabled by default.
- `max_tokens` is a hard limit covering thinking plus visible response text.
- Anthropic lists Fast mode at $10 input and $50 output per million tokens.
- AWS separately confirms availability through its platform.
Anthropic describes Opus 5 as suited to complex coding, agentic work, and knowledge work. Those are Anthropic’s product claims, not independent proof that the model will be the best choice for every task.
Same listed price does not mean the same bill
A token rate is not the same thing as a workflow cost.
Anthropic says requests that ran without thinking on Opus 4.8 will run with thinking enabled on Opus 5. For difficult, multi-step work, that may be helpful. For routine tasks such as ticket classification, product-data formatting, meeting-note summaries, or standard follow-up drafts, it may introduce behavior a workflow does not need.
The key implementation detail is `max_tokens`. Anthropic documents it as a hard cap covering both reasoning tokens and visible output. A workflow tuned for a prior model may need a different output budget, effort setting, prompt format, or fallback rule.
Do not decide based on one impressive prompt. Run a controlled test on 10 to 20 real examples and compare:
- Quality and error rate.
- Time to a usable answer.
- Input and output token use.
- Whether answers become longer than the workflow needs.
- Whether requests hit the `max_tokens` limit.
- Whether parsers, automations, and downstream tools still work.
Keep the existing production model available until the results are clear.
The honest takeaway: Anthropic Opus 5 is worth a controlled test because the listed base API rate matches Opus 4.8. But default thinking can change token budgets and production behavior. Test first; switch second.
Bottom Line
Opus 5's unchanged listed API price does not make migration automatic: default thinking can change token use and behavior, so teams should test production workflows before switching.
