Rumored Kimi-K3: Leaks and Benchmark Roundup
智东西 · wechat · 2026-07-16
Suspected leaks and tests regarding Kimi-K3 are gaining traction. This article compiles multiple rounds of external testing on the anonymous model Kivine, widely speculated to be the upcoming Kimi-K3 from Moonshot AI. K3's performance in frontend generation, 3D interaction, and gaming tasks is considered highly impressive, with some claiming it feels "wilder and more complex" than the Opus and GPT series.
The article also aggregates rumored specs: K3 might feature 2–3T total parameters, a 1M context window, and a novel KimiDeltaAttention (KDA) architecture optimized for long-horizon Agent tasks. Other sources suggest it might combine MLA and support Agent Swarm, allowing complex tasks to be split across multiple sub-agents for parallel execution. The author concludes that while K3 might not immediately surpass Anthropic/OpenAI flagships, matching or exceeding Opus 4.8 in coding would mark a significant breakthrough for domestic models.
Related event: Kimi K3 hype builds as KIVINE appears on Arena(43 posts)→
More from coding & agent
- SF 'Code Engineering' event picks up where 1,000-person harness engineering meetup left off — LangChain · 2026-10-27
- Prompt turns Claude Code into an iPhone backup forensics tool exposing 30-day 'deleted' photo copies — alex_verem · 2026-10-27
- Prompt turns Claude Code into a forensics tool finding deleted iPhone photos in backups — alex_verem · 2026-10-27
- Phones won't be mobile IDEs but control planes for agent fleets, argues Yuchen Jin — Yuchenj_UW · 2026-10-27
- FrontierCS 2 launches on Kaggle: 48 tasks make agents beat published CS papers — MeganRisdal · 2026-10-27
- Stop remaking old games with AI: a call for more ambitious agentic software — rschu · 2026-10-27