Anthropic
Claude Opus 4.5 pricing
Previous-generation Opus. Matches 4.6 on price but with a 200K context window. Pin to this version only when you've benchmarked against it.
Input
$5.00/ 1M tok
Output
$25.00/ 1M tok
- Context window
- 200K
- Max output
- 64K
- Cached input
- -
- Verified
- 2026-04-06
Opus 4.5 is the direct predecessor to 4.6. Pricing is identical - $5 per 1M input, $25 per 1M output - but the context window is 200K instead of 1M, and the max output tops out at 64K. For most new deployments there's no reason to pick 4.5 over 4.6: same money, less capability.
The exception is version pinning. If you've benchmarked a workflow specifically against Opus 4.5 and need to reproduce those results exactly, stay on this version until you can re-benchmark. Anthropic keeps older Claude models available for long enough that this is a reasonable position for established production code.
Calcis counts input tokens against Anthropic's countTokens API for the exact figure Anthropic uses to bill, and predicts output length from the prompt itself.
Estimate your cost on Claude Opus 4.5
Paste your prompt into the estimator, pick Claude Opus 4.5, and see the exact dollar cost - input tokens counted with the provider's own tokenizer, output tokens predicted by our regression model.
Frequently asked
- Should I use Opus 4.5 or 4.6?
- Pick 4.6 unless you have a specific reason to stay on 4.5. Same price, larger context window (1M vs 200K), larger max output (128K vs 64K). The only reason to stay on 4.5 is version pinning for reproducibility.
- How much does Opus 4.5 cost per request?
- A 1,000-token prompt with a 500-token reply costs about $0.0175 ($0.005 input + $0.0125 output). Same as Opus 4.6 - the capability is the differentiator, not the price.
- Does Opus 4.5 support prompt caching?
- Yes, Anthropic's prompt caching works on Opus 4.5 the same way it does on 4.6. Cache writes cost 1.25x the input rate; cache reads cost 0.1x (a 90% discount).
- What tokenizer does Opus 4.5 use?
- Anthropic doesn't publish the tokenizer. Calcis uses Anthropic's own countTokens API for the authoritative count - the same boundary Anthropic bills against.
Pricing verified 2026-04-06 from the provider's rate card.