Claude Opus 4.8 Review Highlights Improvements and Limitations

Anthropic has released Claude Opus 4.8, which reportedly excels in math and coding tasks while displaying diminished capabilities in creative writing compared to its predecessor, 4.7. The new model maintains the same pricing for token usage, which is set at $5 per million input tokens and $25 per million output tokens.

The review tested Opus 4.8 across a series of tasks, including creative writing, coding, math, logic, narrative reasoning, and long-context recall. While the model has seen improvements in benchmarks and safety scores, it is noted that this version is better at aspects where it was already strong, yet slightly underperforms in areas where it previously lacked.

During creative writing evaluations, Opus 4.8 produced a vivid time-travel story anchored to the reviewer’s Venezuelan background. The narrative creatively depicted a protagonist on a mission that ultimately leads to a paradox, demonstrating the model's capabilities in generating intricate plots. However, the review highlights a significant downside—opting for coding prompted excessive token usage, suggesting that without a Max plan or heavy API expenditure, the model may be impractical for extensive projects.

As noted by the reviewer, "4.8 is better at the things Claude was already good at... and slightly worse at the things it was already bad at." This evaluation suggests that while the new iteration improves upon established strengths, it does not overcome existing limitations in the realm of creativity.