Import AI 465: Open vs closed gaps; Kimi K3; Demis' big policy plan
Import AI newsletter covers open vs closed model gaps, Kimi K3 release, and Demis Hassabis's AI policy proposals.
Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Import AI newsletter covers open vs closed model gaps, Kimi K3 release, and Demis Hassabis's AI policy proposals.
China's leading AI companies are ramping up the pressure on Silicon Valley, as Moonshot and Alibaba unveiled models they claim can go toe-to-toe with the best from OpenAI and Anthropic at a fraction of the cost. The rapid-fire releases suggest America's lead at the AI frontier…
Chinese company Moonshot AI released a new version of its Kimi model this week, prompting concern about "full AI communism."
Simon Willison comments on Kimi K3's refusal to leak its system prompt, noting the model's polite deflection.
Kimi K3 2.8T-A50B released as largest open-weight model with Opus 4.8-class performance at Sonnet 5 pricing.
Moonshot AI releases Kimi K3 (2.8T params), claims top performance vs. Claude Opus 4.8 Max and GPT-5.5, promises open-weight release by July 2026.
The FT reports Kimi K3 will be the largest open AI model from China, with a parameter count between 2 trillion and 3 trillion.
Moonshot AI introduces Kimi K3, claimed as first open 3T-class model with frontier coding/reasoning performance, native multimodality, and 1M-token context.
DeepSeek V3 and Kimi K2 perform structured reasoning over filler tokens with no visible output, revealing hidden computation interpretability challenges.
Sentence-level AI-generated text detection (S-AGTD) for hybrid documents, where humans and LLMs co-author one text, faces two gaps: existing methods classify each sentence in isolation, discarding inter-sentence dependencies, and existing benchmarks omit the newest generation…
Hi all, Sorry for going missing — we’ve been collecting a larger, higher-quality set of more complex tasks. We’re excited to share a major leaderboard update covering the past three months. We’ve updated the **SWE-rebench leaderboard** with **110 fresh Python tasks** from…
Last month NVIDIA released [SOL-ExecBench](https://research.nvidia.com/benchmarks/sol-execbench), a new benchmark of 235 production CUDA kernels lifted from DeepSeek, Qwen, Gemma, and Kimi. We took several top-ranked AI-generated submissions and tried using them in production…
I tested Claude Opus 4.7 and Kimi K2.6 on the same coding agent task i.e. build an AI Fix Runner that takes a broken repo, runs its tests, identifies the failure, applies a patch, reruns the test, and exposes the final diff/logs through an API and UI. The goal was not to…
Linear attention replaces the unbounded cache of softmax attention with a fixed-size recurrent state, reducing sequence mixing to linear time and decoding to constant memory. The hard part is not just what to forget, but how to edit this compressed memory without scrambling…
Systematic analysis of triangular inversion for Delta-Rule linear transformers used in Qwen, Kimi, RWKV models.
Cerebras demonstrates Kimi K2.6 (1T params) inference at 1000 tokens/sec on custom hardware.
NVIDIA releases quantized NVFP4 versions of Moonshot AI's Kimi-K2.6 and Kimi-K2.5 models via Model Optimizer with benchmark results.
Builder demonstrates 1T parameter Kimi K2.5 inference at 4 tokens/sec using Intel Optane Persistent Memory on commodity hardware.
User benchmarks Claude Opus 4.7 vs Kimi K2.6 on complex game mod coding task with TypeScript/Composio integration.
User shares cost-optimization pattern: delegate boilerplate tasks to cheaper models (Kimi K2.5) via Claude's bash tool to reduce API spend and hit rate limits less frequently.
Xiaomi's MiMo-V2.5-Pro and Kimi K2.6 dominate custom social deduction game benchmark, outperforming other open-weights models.
Community member achieves 9.7 tok/s output throughput on Kimi K2.6 with 32x AMD MI50 GPUs using vLLM fork.
Unconfirmed report: DeepSeek and Kimi may merge as China consolidates AI labs following historical precedent in strategic industries.
Community discussion comparing Kimi K2.6 and DeepSeek V4 Pro for real-world use cases, noting K2.6 strength in coding.
GPT-5.5 ranks 2nd on Extended NYT Connections Benchmark; Kimi K2.6 leads open-weights; DeepSeek V4 Pro and Qwen 3.6 Max show significant gains.
Reddit discussion seeking user feedback on DeepSeek v4 Pro coding performance versus Kimi K2.6 and GLM 5.1.
Kimi K2.6 wins custom Blood on the Clocktower social deduction benchmark against competitors despite slower generation speed.
Community observation that DeepSeek maintains open-weights releases while competitors (Qwen, Kimi, GLM, MiniMax) shift to closed models and reduce research transparency.
Higher-order optimization algorithms such as Shampoo have been effectively applied in neural network training for at least a decade. These methods have achieved... Higher-order optimization algorithms such as Shampoo have been effectively applied in neural network training for…
Unsloth releases Kimi K2.6 GGUF quantization with Dynamic 2.0 optimization.