Cohere

Discussion of Cohere over the past four weeks was dominated by frustration, with bug reports emerging as the single largest theme by a wide margin. Commenters raised repeated concerns about billing and usage tracking failures, particularly around rerank endpoints where costs appeared as null or zero in logs and interfaces. Several mentions described these not as minor annoyances but as silent failures that distort cost visibility in production environments. The reliability theme reinforced this mood, with discussion suggesting that core functionality was behaving unpredictably across different deployment contexts.

The score trajectory tells a clear story of accelerating decline. Early in the window scores sat in the high sixties to mid-seventies, with only a handful of mentions, suggesting a relatively calm baseline. As mention volume surged sharply in early to mid June, sentiment dropped with it, pointing to a concentrated wave of negative reports rather than a slow drift. The combination of low scores and high mention counts in the final two tracked periods indicates that the most recent weeks captured a genuine spike in dissatisfied voices rather than statistical noise from a small sample.

Praise themes were present but modest in comparison. Feature praise and integration positives did appear, and some commenters acknowledged new releases and areas of solid performance. However, these positive signals were numerically outweighed by complaints, and the feature-missing and integration-lacking themes suggested that even users who found value in the product were bumping against gaps.

Opinion was divided most visibly around integrations. A portion of discussion flagged good integration experiences while a roughly comparable count described integrations as lacking or broken, hinting that outcomes varied considerably by use case or deployment path. Documentation inconsistencies, such as mentions of mismatched model names between front-end documentation and back-end configuration, added a layer of confusion that colored how commenters interpreted other issues, making it harder to separate product problems from setup problems.