Kimi K3 first open model in ErrataBench Top 10
a day ago
- Kimi K3 is the first open weights model to score over 90% on the ErrataBench benchmark.
- It outperforms Opus 4.8 in both performance and price, approaching 5.6 Terra-level performance.
- For local models with sufficient hardware, Kimi K3 is the best choice for English writing and editing tasks.
- ErrataBench evaluates models' ability to detect errors in large text samples and apply valid fixes using tools.
- This milestone marks the first time an open model has entered the top 10 of the benchmark, previously dominated by proprietary models.
- Note: Kimi K3 is described as a 'soon-to-be-open' model.