Claude Opus 4.7
- Entity ID:
ent-20260423-r30a00000000e - Type:
product - Scope:
private - Status:
active
Description
Anthropic model released April 15, 2026. New tokenizer producing ~1.0-1.35x more tokens for the same text (language- and code-mix dependent), better SWE/HumanEval coding benchmarks (~'one step better' than 4.6), improved long-context retention above 200K tokens, more accurate and consistent visible thinking. Pricing unchanged at $5/$25 per million input/output tokens, so worst-case effective cost per kilobyte rises ~35%.
Key claims
- Opus 4.7 tokenizer produces 1.0-1.35x more tokens per text
- Opus 4.7 hits autocompact threshold earlier because of the new tokenizer
- xhigh sits above high in the Opus 4.7 effort hierarchy
- Opus 4.7 is 'literally one step better' on SWE benchmarks
- Opus 4.7 better retains long context above 200K tokens
Relations
- Opus 4.7 Tokenizer --[contains]--> Claude Opus 4.7
- xhigh Effort Level --[contains]--> Claude Opus 4.7