Claude Opus 4.8 Review: Better at What It's Good At, Worse at What It's Not

Anthropic has released Claude Opus 4.8, six weeks after the launch of Opus 4.7. According to a detailed review of the model, the update delivers higher benchmarks and improved safety scores, but the pricing remains fixed at $5 per million input tokens and $25 per million output tokens.

The update is a refinement of existing strengths rather than a fundamental shift in capability. Opus 4.8 shows clear wins in math and produced the cleanest one-prompt game tested. However, the model struggles with the areas where its predecessors were already weak, such as imagination and creative writing.

The performance delta between 4.7 and 4.8 suggests a narrowing of the frontier. While the model excels at mechanical tasks, its creative prose lacks the fluidity and momentum found in models like MiMo v2.5. In testing, the model's writing was descriptive and competent, yet it failed to match the surprises or interesting narrative depth of competing models.

The most significant risk for users is the model's efficiency. A single coding prompt was enough to drain an entire Pro token quota. This makes Opus 4.8 impractical for large-scale projects unless users move to a Max plan or commit to heavy API spend.

The release demonstrates a common trend in frontier model development: optimizing for logic and precision at the expense of resource efficiency and creative nuance. For developers, the choice is no longer just about intelligence, but about the cost of the token appetite.

Evaluate your current token consumption before migrating large-scale coding workflows to 4.8.