Search AI Products and News

Kuaishou opensources the KAT-V1 autonomous thinking large model, which includes two versions: 40B and 200B. The 40B version performance is close to DeepSeek-R1, and the 200B version outperforms several flagship models. The model innovatively adopts a mixed training paradigm of short and long-term thinking and the Step-SRPO reinforcement learning algorithm, which can automatically adjust the thinking mode based on the complexity of the question, solving the problem of overthinking. Based on Qwen2.5-32B, it achieves excellent performance in fields such as science and code through an heterogeneous distillation framework and pre-training with 10 million examples.

Brazilian programmer won $50K in K Prize AI contest with only 7.5% accuracy, revealing AI model limits. $1M prize offered for 90% accuracy in GitHub-based test.....

The Kimi team released the Kimi K2 technical report. This trillion-parameter large model, with 32 billion activated parameters, won the global open-source model competition, with performance comparable to top closed-source models such as GPT4.5. Its innovations include: using the MuonClip optimizer to process 15.5 trillion token data without loss; developing a multi-domain tool data synthesis pipeline; pioneering the restatement method to improve training efficiency, achieving the effect of ten rounds of original data in one round of rewritten data; optimizing model performance through supervised fine-tuning and reinforcement learning. The model is based on N

NTU's 'WORLDMEM' model enhances virtual reality with long-term scene consistency via innovative memory mechanisms and conditional diffusion transformers.....

In 2025, open-source large models will show three major trends: 1) MoE architecture becomes mainstream, with DeepSeek-V3 (67.1 billion parameters) and Qwen3-235B (235 billion parameters) each having unique designs in their expert systems; 2) small models break through performance bottlenecks, with SmolLM3-3B adopting position-encoding-free technology, and Qwen3-4B achieving lightweight efficiency; 3) models show significant differentiation, with Llama3.2 focusing on general tasks, while Kimi-K2 (1 trillion parameters) excels in complex reasoning.

Moonshot AI's open-source model Kimi K2 excels in programming, matching Qwen3-235B-A22B and nearing o3-mini-high & Claude-3.7-Sonnet. With 1T MoE, 128k context, 65.8% accuracy on SWE-bench, and $0.14/M tokens, it's ideal for coding. Supports web generation, workflows, vLLM/Hugging Face deployment, MIT-licensed.....

Cursor's regional restrictions drive developers to adopt Chinese AI Kimi K2, hitting 10B daily tokens. Its API is 5-25x cheaper than Claude, gaining global traction in coding and content creation.....

Moon's Dark Side addresses Kimi K2API slowdowns, citing high traffic and large model size. Optimization and hardware upgrades underway, with improvements expected soon. Kimi K2 is fully open-source, allowing alternative deployments. The company, founded in 2023, focuses on AI tools like Kimi Assistant for translation and legal analysis, aiming to enhance user experience.....

Kimi K2 excels in creative writing, outperforming o3-Pro in short story creation. This open-source model by Moonshot (Ali-backed) shows strengths in literary compression and metaphor innovation, with some works near publishable quality. Its low cost ($0.15/M tokens) and precise instruction-following attract developers, though emotional depth and multilingual performance need improvement. This breakthrough sets new AI writing standards.....

TRAE.ai launches Kimi-K2 model and Grok-4(Beta). Kimi-K2 excels in code/math with MoE architecture, rivaling GPT-4.1. Easy 3-step access. International version adds Grok-4(Beta) testing alongside Claude, Gemini, GPT.....

On July 11, Moonshot officially launched the Kimi K2 model, which has stronger code capabilities and general agent task processing abilities, and simultaneously chose to open-source it. This foundation model based on the MoE architecture has a total parameter count of 1T and an activated parameter count of 32B, and it immediately attracted widespread attention upon its release.​ However, recently some users have reported that the API service speed for the Kimi K2 model is slow. In response to this, Moonshot posted tonight stating that the main reasons for the slow speed are the high volume of traffic and the large model size. To address this issue, the company

Moonshot AI's Kimi K2, a 1 trillion-parameter open-source model, surpasses Grok in market share. It excels in 128K context reasoning, outperforms Claude/GPT-4 in benchmarks, and offers free API. Now open on Hugging Face.....

Unsloth AI quantized Moonshot AI's 1T-parameter Kimi K2 model to 1.8bit, reducing size by 80% to 245GB while maintaining performance. The MoE-based model excels in coding and reasoning, now deployable on 512GB M3Ultra devices, lowering costs. This advancement positions Kimi K2 as a GPT-4.1 competitor, benefiting SMEs and boosting open-source AI adoption in education/healthcare.....

Moonshot AI's open-source model Kimi K2, with 1 trillion parameters, excels in coding and reasoning. Its low cost and open strategy quickly gained 1.5% market share, surpassing Grok4 and GPT-4.1.....

The Silicon Base Flow platform has launched the open-source MoE model Kimi K2 developed by Moonshot AI. The model has a total of 1T parameters and 32B activated parameters, supports a context length of 128K, and performs excellently in coding and agent tasks. The pricing is 4 yuan per million tokens for input and 16 yuan per million tokens for output. New users can get 14 yuan in trial credit upon registration. The model has three technical advantages: 15.5T tokens of large-scale training, MuonClip optimizer for stable expansion, and design optimized for agent tasks. Tests show that it excels in coding

The CEO of American AI company Perplexity publicly praised the newly released Kimi K2 model with one trillion parameters from Moonshot AI, stating that they may conduct training based on this model. Kimi K2 uses a mixture-of-experts architecture, showing excellent performance in tasks such as code generation, intelligent agents, and mathematical reasoning, and has achieved leading results in multiple benchmark tests. Perplexity previously used the DeepSeek R1 model, and this statement shows that they are actively seeking technological collaborations to drive AI product development. Kimi K

MoonDark releases trillion-parameter open-source model Kimi K2, featuring MoE with 32B active params. Excels in coding/math, supports tool use & code execution. Uses MuonClip optimizer, offers API & base model. Key competition to closed-source AI, available via official platforms.....

Moonshot AI officially released its latest creation - the Kimi K2 model, and simultaneously announced its open source. This foundation model based on the MoE architecture has gained widespread attention in the AI field since its release, thanks to its strong coding capabilities and excellent general Agent task processing abilities. The Kimi K2 model has a total of 1T parameters, with 32B activated parameters. It has achieved top performance among open-source models in a series of benchmark performance tests such as SWE Bench Verified, Tau2, and AceBench.

Kunlun releases Skywork-R1V3.0, an open-source multimodal model achieving expert-level reasoning. It excels in physics (52.8), logic (59.7), and math (77.1), outperforming Claude-3.7-Sonnet and GPT-4.5 with only 25K training samples.....