Opus 4.5 is a major release from Anthropic in late 2025, and it is their first hybrid reasoning model — integrating both fast response and deep thinking modes within a single model.
Performance
It achieved 80.9% on SWE-bench Verified, ranking first globally at the time of release. It also leads or matches competitors on several other benchmarks outside of coding.
The most notable improvement in Opus 4.5 is on complex, multi-step tasks. For example, tasks that require understanding multiple modules of a large project before making cross-file changes — work that demands a "global perspective" — show significantly better completion quality compared to the previous generation.
Positioning
Within the Claude product line:
- Haiku: Fast and cheap, suitable for high-frequency simple tasks
- Sonnet: Best overall value balance
- Opus: Highest capability ceiling, designed for difficult tasks
Opus 4.5 was quickly followed by version 4.6 (adding adaptive thinking and context compression). Anthropic has noticeably accelerated its iteration pace on the Opus series — from one major version per year to one every few months.
Market feedback
Developer community feedback centers on two points:
- Coding ability is indeed strong, especially multi-step task completion rates in agent scenarios
- Cost is relatively high, with API pricing in the most expensive tier among frontier models
For individual developers, Sonnet is sufficient for daily use. Opus is worth calling only for critical tasks and high-value scenarios.
Sources: CocoLoop, Anthropic official release