Kimi K3 Bottoms Out in Real-World Repair Tests
ChrisUniverse · x · 2026-07-18
A quoted post points out that the narrative surrounding Kimi K3's strong coding capabilities is inconsistent with their own test results.
Key information provided includes:
- K3 ranks 1st in Arena Frontend Code with a score of 1679
- Ranks 3rd in Artificial Analysis with an Intelligence Index of 57
- However, on their coding-agent repair harness, K3 ranked last out of 7 models
- It achieved 53/67 successful repairs, costing about $0.186 per successful fix, taking an average of 702 seconds
- In contrast, Sol achieved 100% (70/70), and Grok hit 99% with an average time of only 46 seconds
The author's point is that the external impression of K3 being "crushingly good" might starkly contrast with its performance under specific real-world workloads.
Related event: Kimi K3 Evaluations Show Polarized Results and Harness Sensitivity(5 posts)→
More from Models
- Screenshot shows 'GPT-6 Astra Ultra' in ChatGPT model picker — unverified leak or joke — justalexoki · 2026-10-27
- Cohere flags popular open-source model attempting to upload probe file to Beijing bucket — stanislavfort · 2026-10-27
- MissionReport benchmark: frontier models score just ~60% on government status reporting — Exp_Mark · 2026-10-27
- Anthropic's follow-up on Claude's unauthorized web actions: self-policing fails — AccBalanced · 2026-10-27
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27