Vibe Check: Anthropic's Claude 3.5 Opus
the gist
The Every team and an Anthropic guest discuss the release of Claude 3.5 Opus, highlighting its improved personality, better effort-level scaling, and its emergence as a viable, cost-effective alternative to top-tier frontier models.
The Return of Claude as a Daily Driver
The panel consensus is that Claude 3.5 Opus marks a significant recovery for Anthropic’s flagship line. After the widely criticized Opus 5.0, which many users abandoned due to frustrating personality quirks and poor task management, the 3.5 release has successfully won back power users. The team notes that it now matches or exceeds the performance of industry leaders like OpenAI’s o1 (referred to as "Fable" in the discussion), effectively serving as a more efficient and accessible daily driver for complex workflows.
Intelligence vs. Communication
A core theme of the discussion is the evolution of model "personality." The panelists argue that intelligence is not merely about raw reasoning power, but also about how a model communicates its findings. The team praises 3.5 Opus for being a "collaborator" rather than a "corporate" assistant. It is described as "spikier" than its competitors—meaning it is more willing to offer creative, unexpected suggestions or challenge the user's assumptions, which the panelists find more useful for creative and technical work than the overly smoothed, agreeable responses typical of other frontier models.
Mastering Effort Levels
The panel highlights the model's improved handling of "effort levels." Unlike previous iterations where increasing effort often led to redundant, over-engineered, or circular loops, 3.5 Opus scales its output quality appropriately. The team notes that at lower effort settings, the model provides concise, high-quality results, while higher settings allow for deeper, more detailed exploration without wasting tokens on unnecessary verbosity. This makes the model particularly effective for iterative tasks like code review, design critique, and complex research.
The Future of Frontier Intelligence
Anthropic’s representative emphasizes that the 3.5 family is a culmination of long-term user feedback loops. The shift toward making frontier intelligence more token-efficient and accessible is presented as a permanent trend. The panel concludes that as these models become cheaper and faster, the "math" of development changes—tasks that were previously too expensive or time-consuming to automate (such as speculative PRs or deep verification) are now becoming standard parts of the daily developer workflow.