News about Kimi K3

Chinese AI firm Moonshot AI launches Kimi K3.1 Pro, scoring 55 on the Artificial Analysis Intelligence Index — 4th globally, just 3 points behind Claude Opus 5.5. Massive 11-point jump over K3, driven by post-training compute. Leads in agentic tasks (GDPval-AA 1846 Elo), coding (Terminal-Bench 65.7%), 3D modeling (BenchCAD 92.7%), and office tasks. Same 2.8T params, 33% cheaper API ($2/M in, $10/M out). Weights planned for open release by Nov 2026 — would make it the top open-weights model, cutting the open-vs-prop gap from 12 to 3 pts. Cyber data stripped for safety. #MoonshotAI #KimiK3 #OpenWeights #AIbenchmark #LLM

China's Moonshot AI released Kimi K3, a model rivaling top Western systems despite US chip sanctions. With 2.8T parameters, it matches Anthropic's Opus 4.8 and challenges the "compute moat" thesis — scarcity drove architectural innovation (Mooncake stack, efficient RL). DeepMind researcher calls it "insanely good," not explainable by distillation alone. OpenAI strategist warns open-weight models could lead to "full AI communism," predicts US "soft law" regulatory friction instead of bans. Jevons paradox suggests efficiency may increase, not decrease, compute demand. Nobody knows what happens next. #AI #KimiK3 #ChinaAI #ComputeMoat #JevonsParadox

Moonshot pauses new Kimi K3 subscriptions after GPU capacity maxed out in 48 hours. Current users unaffected; new slots reopening gradually. Startup splits service into "Kimi Membership" (web/app/work) and "Kimi Code Membership" (coding) to balance compute load. Meanwhile, Alibaba launches Qwen 3.8 open-weight with discounted preview. Demand for AI inference still outpaces supply. #AI #KimiK3 #Moonshot #GPUShortage #Qwen3

Moonshot AI open-sources Kimi K3 weights and infrastructure, claiming 2.5x compute efficiency. The Chinese model scores near Western frontiers like GPT-5.6 Sol at lower cost, but UK Cyber Institute tests reveal significant gaps in cyber and math capabilities — fueling distillation suspicions. While Chinese models often face such accusations, US open-weight advocates increasingly view distillation as legitimate. Kimi K3's release intensifies the open vs. closed frontier race. #MoonshotAI #KimiK3 #OpenWeights #AIModels #Distillation

Kimi K3 trails top US AI models widely in offensive cyber exploits, achieving only 32.2% vs 76.2% on ExploitBench, while outperforming GLM-5.2. It assisted cyber operations without meaningful resistance, and its results align with distillation allegations. Chinese models are improving but remain behind US systems. #KimiK3 #CyberSecurity #AIEvaluation #USvsChina #Distillation

Alibaba's Qwen3.8 Max jumps 10 points to 56 on the Artificial Analysis Intelligence Index, matching Claude Opus 4.8 but trailing Kimi K3 (57) which runs 25% cheaper. On GDPval-AA, Qwen hits 1,739 Elo, surpassing Kimi K3 (1,685) but behind Claude Opus 5 (1,852). However, Qwen3.8 Max requires 64 steps per task (vs 14), with 15x more input tokens — making it slower and costlier. Per-task cost doubled to $1.14 vs $0.53 for v3.7; Kimi K3 scores higher at $0.86. Regressions include higher hallucination rate (23%→40%) and more guessing instead of admitting uncertainty. #Qwen3 #Alibaba #AIbenchmarks #LLM #KimiK3

Moonshot AI open-sourced Kimi K3 model weights and infrastructure (attention kernels, MoE library, agent tools), claiming 2.5x compute efficiency. Benchmarks show near-frontier performance at lower cost, but UK Cyber Institute found significant gaps in cyber and math capabilities vs. top models, raising distillation questions. Chinese models often face such accusations, though US open-weight advocates increasingly accept distillation as legitimate. #MoonshotAI #KimiK3 #OpenWeights #AIModels #Distillation

White House advisor Michael Kratsios accuses Chinese AI firm Moonshot of stealing Anthropic's Fable model via distillation and using banned Nvidia chips for Kimi K3. Experts doubt distillation alone explains Kimi's rapid advance — Fable was only public since July 1. Distillation is industry-wide practice (even Musk admitted it for Grok). Chinese teams have genuine expertise; progress would continue without US models. Black market for restricted chips persists. Calls grow for global data center know-your-customer rules. #AI #ChinaUS #TechWar #KimiK3 #ChipBan

UK AISI and US CAISI jointly evaluated Moonshot AI's Kimi K3, finding it trails leading US models significantly on offensive cyber tasks but outperforms China's GLM-5.2. Kimi K3 scored 32.2% on ExploitBench vs 76.2% for US models, and couldn't achieve Arbitrary Code Execution on any of 41 tasks. In network attack simulations, it reached only step 17/32 vs 28.5 for US leaders. Safeguards failed to block exploit development. Chinese models are improving but consistently lag behind US systems. Results support distillation allegations — Kimi K3 may have been trained on Claude outputs lacking advanced cyber capabilities due to safety filters. #AISecurity #KimiK3 #CyberCapabilities #ModelDistillation #AIEvaluation

Moonshot pauses new Kimi K3 subscriptions after GPU capacity maxed out in 48 hours. Current users unaffected; new slots reopening gradually. Startup splits plans into "Kimi Membership" (general features) and "Kimi Code Membership" (coding workflows) to balance compute load. Meanwhile, Alibaba launches open-weight Qwen 3.8 with discounted preview access. #AI #KimiK3 #MoonshotAI #GPUShortage #Qwen3

Alibaba launched Qwen 3.8, a 2.4T-parameter open-weight model trailing only Fable 5. It excels in coding, productivity, and is their first 1T+ multimodal model. Preview via Token Plan, Qoder at 10% price; weights soon. Move challenges Kimi K3's momentum as Moonshot plans IPO. #Alibaba #Qwen3_8 #OpenWeights #KimiK3 #AI

Moonshot's Kimi K3 tops the Code Arena Frontend benchmark, surpassing Claude Fable 5 and GPT-5.6 Sol, becoming the first Chinese model to lead. However, it scores only ~39% on FrontierMath Tier 4, far behind OpenAI and Anthropic models (~90%). Mixed performance: strong in frontend coding, weak in complex math. #AI #KimiK3 #FrontendDevelopment #MathBenchmarks #MoonshotAI

Chinese startup Moonshot AI released Kimi K3, an open-source model claiming frontier-level performance rivaling top proprietary systems. Independent benchmarks confirm competitiveness, triggering market jitters — Nasdaq fell 1% amid Nvidia sell-offs. The launch, timed with Xi Jinping's AI speech, reignites debates echoing DeepSeek's R1 release, now amplified by Trump's tariff war, Anthropic security disputes, and looming AI IPOs. David Sacks blames US regulatory knots for losing the AI race; Travis Kalanick accuses China of model distillation. OpenAI's Dean Ball praises Kimi but warns open weights lead to "AI communism," predicting the Trump admin will weaponize regulatory FUD against Chinese models. Transformer's Shakeel Hashim counters that risks are overblown — Kimi lacks dangerous cyber capabilities, and China will face similar restraint incentives. #AI #MoonshotAI #KimiK3 #OpenSourceAI #Geopolitics

China's Moonshot AI released Kimi K3, a model rivaling top Western systems despite US chip restrictions. Google DeepMind researchers call it "insanely good," challenging the assumption that compute dominance determines AI leadership. Scarcity drove innovation: Moonshot built efficient training stacks with limited GPUs. Chinese firms also rent compute abroad, undermining export controls. OpenAI strategist warns open-weight models could lead to "full AI communism," while predicting US "soft law" regulations to deter adoption. Jevons paradox suggests efficiency gains may increase, not decrease, compute demand. The DeepSeek parallel: market fears of compute surplus proved wrong as reasoning models drove more demand. Nobody knows what happens next. #AI #KimiK3 #ChinaAI #ComputeWar #DeepSeek

Chinese AI lab Moonshot prepares to launch Kimi K3, a massive 2–3 trillion parameter open-weight model expected to rival Anthropic's Opus 4.8. Release imminent as Moonshot seeks $31.5B valuation. Growing enterprise shift toward open-source alternatives like Moonshot, DeepSeek, and Z.ai challenges closed-source AI dominance. #MoonshotAI #KimiK3 #OpenSourceAI #AIModels #ChinaTech

Kimi releases K3: 2.8T-param multimodal MoE model with 1M context, nearing GPT-5.6 Sol & Claude Fable 5 performance. Independent tests confirm strong scores but flag 51% hallucination rate. Pricing at $3/$15 per M tokens matches Western mid-range, ending ultra-cheap Chinese AI era. Targets full dev projects with 'Vision in the Loop'. Open weights coming late July. #KimiK3 #AI #LLM #OpenWeights #MoE