Kimi K3 Review: The New Open-Source King with Stellar Search
andersonbcdefg · x · 2026-07-19
Independent reviewers added Kimi K3 to the prinzbench benchmark. K3 achieved an overall score of 47/99, far surpassing GLM-5.2's 30/99, making it the best open-source model tested to date.
- Overall Comparison: Its score is slightly below February's Gemini 3.1 Pro (50/99), putting them in the same tier. K3 excels slightly in search capabilities but is somewhat weaker in logical reasoning.
- Stability: K3 shows notable volatility; it occasionally misses simpler questions but frequently delivers stunning results on highly difficult ones.
- Core Highlight: With a search sub-score of 11/24, it becomes the first non-OpenAI model to correctly answer more than 10 search questions in this benchmark.
More from Models
- Fangcun AI, China's first independent frontier AI safety lab, launches with DeepSeek eval report — JacquesThibs · 2026-10-26
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26