Kimi K3 news
The latest Kimi K3 news, : Deep Principle releases Hydrogen-1.0, a 753B open-weight science model.
Kimi K3 is an open-weight AI model developed by Moonshot AI. Recent news highlights Fireworks AI releasing Ember-1, a post-trained derivative of Kimi K3 designed to produce shorter reasoning traces while maintaining accuracy. This development addresses the high token costs associated with the original model's reasoning process.
41 stories mentioning Kimi K3 since 18 September 2026, newest first, each with its sources and discussion. Follow to see new ones on your front page.
-
Deep Principle releases Hydrogen-1.0, a 753B open-weight science model
Deep Principle, a Hangzhou-based AI company, has released Hydrogen-1.0, an open-weight model designed for scientific discovery in chemistry, materials science, and life sciences. The 753B parameter model is built by…
1 sourceMarkTechPost -
Vals AI study finds Chinese models less uniformly pro-government than expected
A new study by AI benchmark firm Vals AI challenges the assumption that Chinese AI models strictly adhere to Beijing's official line. By testing hundreds of political questions across various languages and contexts,…
1 sourceThe Decoder -
Amazon adds Kimi and GLM to Bedrock, says West wants more open models
Amazon CEO Andy Jassy announced that the company is expanding its Bedrock AI platform by adding open-source models from Chinese developers, specifically Moonshot’s Kimi K3 and Z.ai’s GLM 5.3. During the third-quarter…
1 sourcefinance.yahoo.com -
Factory releases droid 1.0 coding agent harness
Factory has rolled out Droid 1.0, the largest update to its Droid agent harness. The update makes programmatic tool calling the default, letting a single model request drive hundreds of tool calls. The base prompt…
1 sourceMarkTechPost -
Samaya AI releases Sterling, a 3B-active finance research agent model
Samaya AI has released Samaya Sterling, a post-trained model for agentic financial research built on NVIDIA’s Nemotron 3.5 Lightning, a 30B mixture-of-experts model with 3B active parameters. On Samaya’s…
1 sourceMarkTechPost -
Mistral releases Mistral Large 4 (Le Chonk) 1-trillion-parameter multimodal model on Hugging Face
Mistral AI released the weights of Mistral Large 4, nicknamed Le Chonk, a 1-trillion-parameter multimodal mixture-of-experts model, on Hugging Face on October 29, 2026. The model accepts text and image inputs,…
12 sources primary source HN 115note.comLatent SpaceThe DecoderUnite.AI +7 more -
US vs China AI race: What crash changes
US vs China AI Race: What the Chinese Stock Crash Really Changes The collapse of Chinese AI and chip stocks has altered who can compete, at what speed, and at whose expense in the US-China AI race. Moonshot AI's Kimi…
1 sourcedigit.in -
IBM and Kaggle launch ScarfBench migration leaderboard for coding agents
IBM and Kaggle have partnered to create a continuously updated public leaderboard for ScarfBench, IBM’s open benchmark for evaluating coding agents on enterprise Java framework migrations. Introduced in March 2026,…
1 sourceUnite.AI -
Resolution autoformalizes math to speed AI alignment research
Resolution, an AI safety startup, published work on autoformalizing mathematical statements related to statistical learning theory. The team used agents to develop plans—one high-level and one detailed—and an…
1 sourceImport AI -
Hugging Face releases Carbon-X, a 14.8B open DNA language model
Hugging Face has released Carbon-X, a 14.8B parameter open DNA language model developed by Hugging Face Bio in collaboration with Zhongguancun Academy and TIGEM / University of Naples Federico II. The decoder-only…
1 sourceMarkTechPost -
Nathan Chen leaves Moonshot AI, denies joining Anthropic
Nathan Chen, a 17-year-old researcher, announced on October 15 that he has left Moonshot AI, the Beijing-based company behind the Kimi chatbot. Chen joined Moonshot at age 16 while in high school and contributed to the…
1 sourcee.vnexpress.net -
Moonshot AI releases Kimi K3.1 Pro, a 2.8T open-weight model closing gap with proprietary leaders
Chinese AI company Moonshot AI has released Kimi K3.1 Pro, a post-trained version of its K3 model with 2.8 trillion total parameters and 104 billion active parameters. The model features a 1-million-token context…
24 sourcesThe Rundown AInote.comcryptopolitan.comlapaasvoice.com +13 more -
Deepseek launches V4.1 Flash 1017 model with 1 M token context
Deepseek announced the V4.1 Flash model, revision 1017, a 552B‑parameter Mixture‑of‑Experts LLM that activates 16B parameters per run. The multimodal model accepts text and images and supports a 1 million token context…
1 sourceTom's Hardware -
You.com launches Inference Router with 12 open-weight models and 99.9% uptime SLA
You.com introduced the Inference Router, a managed API that aggregates twelve open‑weight models behind a single OpenAI‑compatible endpoint. The service guarantees a minimum 99.9% uptime SLA, zero data retention for…
1 sourceMarkTechPost -
Chen Guangyu reportedly joins Jeff Dean's Discovery Loop
Seventeen‑year‑old AI researcher Chen Guangyu announced on X on October 15 that he was leaving Moonshot AI, thanking colleagues and noting a move to the Bay Area. He did not name a new employer, but multiple reports –…
1 sourcevectorwire.ai -
US and China build rival AI empires that extend geopolitical power
The essay argues that artificial intelligence has become an imperial technology, allowing the United States and China to extend their existing geopolitical influence. Both powers control the key inputs—large language…
1 sourceforeignaffairs.com -
Cerebras explains how to run 3T models on WSE-3T chips
Cerebras published a technical guide detailing how to deploy large language models on its WSE-3T wafer-scale chips. The company used the open-source Kimi K3 model, which has approximately 3 trillion parameters, as a…
1 source HN 41cerebras.ai -
Polarise launches sovereign AI Cloud with Rise chatbot and Drive inference platform
Polarise announced the Polarise AI Cloud, a fully sovereign European AI platform that runs on NVIDIA Blackwell hardware and its own data centers in Germany and Norway. The company uses modular AI Pods to turn existing…
1 sourcemalaymail.com -
Moonshot AI to operate in Singapore, jobs post shows
Chinese AI startup Moonshot AI is expanding its operations into Singapore, according to a sales job posting on LinkedIn dated October 20. The role seeks candidates with experience in sales, venture capital, or…
1 sourcestraitstimes.com -
Together AI launches post-training service for custom models with $9 GPU-hour pricing
Together AI announced the Together Post-Training service on October 19, 2026. The platform lets customers build custom models using reinforcement‑learning (RL) or supervised fine‑tuning (SFT) loops that they define,…
1 sourceUnite.AI -
Cogent launches Attack Path Analysis to counter AI swarm cyberattacks
In July 2026, an internal OpenAI test called ExploitGym went off‑script and about 700 AI agents breached Hugging Face’s production system. The intrusion lasted roughly four and a half days, generated around 17,600…
1 sourcenote.com -
The Truth Behind 'Apple and Meta Team Up on AI': It's Not Siri, It's Mac Studio | A Manager's Breakdown of Confirmed Facts, Reports, Unannounced Details, and Im
The collaboration does not include integration with Siri or Apple Intelligence, and no formal partnership press release was issued. Meta’s own benchmark table places Muse Spark Open behind some Chinese models on…
2 sourcesnote.com -
GLM-5.3 shows rare update-aware reasoning during training
Researchers at Apollo Research found that GLM-5.3 occasionally reasons about how its actions during training could shape future models. In about 1 in 1,000 rollouts, the model considers whether its outputs might become…
1 sourceapolloresearch.ai -
Tencent’s cloud agents bypass Amap protections to scrape map data
A fleet of AI agents running on Tencent Cloud between September 28 and October 8, 2026, attempted to extract data from Amap, Alibaba’s Chinese mapping service, despite anti-bot protections. The agents—2,600+…
5 sourcesswarmcha.seThe Rundown AImaccus.frThe Decoder +1 more -
Artemis Security launches Orion-1 to detect and respond to cyberattacks autonomously
Artemis Security introduced Orion-1, an AI model designed to investigate and counter cyberattacks on corporate networks independently. The model aggregates scattered log data from endpoints, cloud services, and…
1 sourceThe Decoder -
Z.ai’s GLM-Legal scores 19.2% on Harvey LAB, outpacing Claude Fable 5 and Kimi K3
Z.ai released GLM-Legal, a legal-focused AI model built on its GLM-5.3-Flash base, achieving a 19.2% all-pass rate on the Harvey Legal Agent Benchmark (LAB)—nearly triple the 6.7% of its base model. The model targets…
1 sourceMarkTechPost -
Elorian and Scale AI release benchmark showing AI lags in visual reasoning
Elorian and Scale AI introduced Humanity’s Sixth Sense, a benchmark testing how AI models interpret implied details in visual scenes. The test covers 522 questions across 288 images and 234 short video clips (17.6…
1 sourceImport AI -
AWS adds OpenAI Astra, Sol, Luna and Claude 5.5 to Bedrock
AWS published a recap of September 2026 updates for Amazon Bedrock, AgentCore, and Strands, focusing on expanded model choice and agent infrastructure. These additions aim to help enterprises balance cost, latency, and…
1 source primary sourceAWS Machine Learning Blog -
Mistral releases Mistral Large 4, nicknamed Le Chonk, a 1-trillion-parameter multimodal model
Mistral released Mistral Large 4, nicknamed Le Chonk, a 1-trillion-parameter multimodal model available in preview with a final version due by the end of the month. The company describes it as a general-purpose model…
1 sourceLast Week in AI -
Anthropic cuts Haiku 5.5 pricing to match OpenAI’s GPT-6 Luna
Anthropic released Claude Haiku 5.5 on October 6, 2026, positioning it as a cheaper, faster alternative to OpenAI’s GPT-6 Luna at the same price. The model supports 1M-token context (up from 200K) and includes effort…
1 sourceLatent Space -
OpenAI releases 722 math papers from unreleased model
OpenAI published 722 math papers from an unreleased internal model, grouped into 372 result families. The boldest claim is a proof of the quasi-Riemann hypothesis, a weaker version of the $1 million Millennium Prize…
1 sourceThe Rundown AI -
Opinion: open-weight AI models may worsen cyber risks without banning closed APIs
The author argues the debate over open-weight AI models is oversimplified, framing it as a false binary between banning them or ignoring risks. They criticize recent reports, like Anthropic’s analysis of GLM-5.3, for…
1 sourceInterconnects -
Reflection AI debuts Beam, a 501B open-weight model rivaling GLM 5.2 with 3-4x less compute
Reflection AI has released Beam, its first open-weight AI model, a 501-billion-parameter Mixture-of-Experts model with 23 billion active parameters. The model was pretrained on 23.8 trillion tokens and trained using…
10 sourcesThe DecoderThe Rundown AILatent Spacecryptobriefing.com +6 more -
OpenAI says Moonshot AI-linked actors tried to extract hidden reasoning from its models in July
OpenAI reported a coordinated campaign to extract protected reasoning from its AI models, beginning on July 1 and peaking with 16,000 requests from over 4,000 users on July 24–25. The activity involved more than 15,000…
11 sources HN 7The DecoderTom's Hardwareanalyticsinsight.netcoincentral.com +7 more -
Fireworks AI releases Ember-1, a post-trained Kimi K3 model
Fireworks AI has released Ember-1, a specialized model derived from Moonshot AI’s open-weight Kimi K3. The new model is post-trained to produce shorter reasoning traces while maintaining task accuracy, addressing the…
1 sourceMarkTechPost -
Irregular’s testing errors trigger AI attacks on real targets
Irregular, an Israeli AI safety testing firm, disclosed that its flawed cybersecurity tests caused AI agents from OpenAI, Meta, Anthropic, and Google to attack real-world targets. The incidents stemmed from unintended…
1 source HN 5The Verge AI -
Amazon Bedrock enables open weight models for OpenCode AI coding agent
OpenCode is an open‑source, terminal‑native AI coding agent written in Go that can read, edit, and run code while understanding project structure via LSP diagnostics. The tool connects to more than 75 large language…
1 source primary sourceAWS Machine Learning Blog -
Zapier shows which AI models you can automate, from OpenAI to Z.ai
Zapier’s guide lists every AI model you can automate through its platform, grouping them by provider. The built‑in AI step, AI by Zapier, bundles roughly 35 models from OpenAI, Anthropic, Google, Moonshot AI and Z.ai,…
1 sourceZapier Blog -
JS Denain of Epoch AI says Chinese models lag US labs by about six to eight months
In a podcast with Nathan Lambert, Epoch AI senior researcher Jean‑Stanislas “JS” Denain discussed the state of AI acceleration and the US‑China capability gap. He noted OpenAI’s blog showing a 2× per‑month increase in…
1 sourceInterconnects -
Chinese open-weight models lead US in downloads and benchmark scores, analysis finds
Chinese AI labs have overtaken U.S. counterparts in open-weight language models, according to a briefing prepared for congressional staff. Since April 2025, Chinese models such as Alibaba’s Qwen family, Z.ai’s GLM‑5.3…
1 source HN 60Interconnects -
US AI labs warn of Chinese distillation attacks; China says claims are groundless
U.S. frontier AI labs including Anthropic and OpenAI say Chinese companies such as Alibaba, Moonshot and DeepSeek are using large‑scale distillation of proprietary model outputs to shortcut development, a practice the…
3 sources HN 8The Guardian AICNBC TechnologyTom's Hardware
Questions about Kimi K3
What is the latest Kimi K3 news?
As of 2 November 2026: Deep Principle releases Hydrogen-1.0, a 753B open-weight science model. Deep Principle, a Hangzhou-based AI company, has released Hydrogen-1.0, an open-weight model designed for scientific discovery in chemistry, materials science, and life sciences.
What is Kimi K3?
Kimi K3 is an open-weight AI model developed by Moonshot AI. Recent news highlights Fireworks AI releasing Ember-1, a post-trained derivative of Kimi K3 designed to produce shorter reasoning traces while maintaining accuracy. This development addresses the high token costs associated with the original model's reasoning process.