Kimi K3

Vercel raises Token Pass allowance to $150 in monthly tokens
Vercel has increased the monthly token allowance included with Token Pass from $80 to $150, while keeping the price at $30 per team per month. The credit applies to…

FAR.AI says simple text attacks bypass Kimi K3.1 Pro safeguards
A few lines of text can bypass Kimi K3.1 Pro’s safeguards on up to 99.4% of tested harmful requests, according to a FAR.AI security stress test. The researchers say the…

Vercel adds GLM 5.3 Flash to its Token Pass subscription
Vercel has added GLM 5.3 Flash to Token Pass, bringing Z.ai’s native multimodal coding model to the subscription’s model lineup. The model is available through Vercel’s…

Mozilla AI launches Otari Subscriptions for open-weight coding models
Mozilla AI is opening early access to Otari Subscriptions, a service designed to make open-weight models easier to use for coding agents. The plans combine multiple…

Together AI introduces compiler that continually optimizes AI inference
Together AI has introduced Together Compiler, a continuous compilation system designed to optimize AI inference as models, hardware and customer traffic change. The…

Kimi K3 gets Indian regional processing on Amazon Bedrock
Amazon Bedrock now offers in-country inference in India for the Kimi K3 model. The option targets local workflows subject to strict data residency requirements, meaning…

Open-weight models hit 72% of top-50 token use on OpenRouter
Open-weight models handled 72% of the tokens among OpenRouter's top 50 models in late September 2026, up from 27% in October 2025, according to a new analysis from the…

Baseten cuts GLM-5.3 delta weight sync from 36 to 15 seconds
Baseten says a delta weight sync for a roughly 770 GB GLM-5.3 checkpoint now takes 15 seconds with a 0.8-second request pause, down from 36 seconds and 6 seconds in its…

DigitalOcean makes prompt caching generally available for Serverless Inference
DigitalOcean has made prompt caching generally available on its Serverless Inference platform, billing cached input at up to 90% off and cutting time to first token by…

CoreWeave Opens Kimi K3 Deployments on GB300 NVL72 Dedicated Inference
CoreWeave customers can now deploy Moonshot AI's Kimi K3 on the company's Dedicated Inference service by bringing their own weights. According to CoreWeave, the model…

Vercel adds DeepSeek V4.1 Flash to its Token Pass subscription
Vercel has added DeepSeek V4.1 Flash to Token Pass, its subscription for Pro teams that runs through AI Gateway. The model joins Kimi K3 and GLM 5.3, and Vercel says…

AMD MI355X Beats NVIDIA B300 per GPU on Kimi K3 AgentX Benchmark
An eight-GPU AMD Instinct MI355X system serving Kimi K3 with vLLM delivers 24.4% more total tokens per second per GPU than NVIDIA HGX B300 at 50 tokens/s/user P90…

vLLM Speeds Up Kimi K3 Decoding on GB300 With Agent-Written Kernels
The vLLM team has published an optimized decode path for Kimi K3 on NVIDIA GB300 GPUs, built with help from coding agents. According to vLLM, it delivers over 840…

Open-weight models now handle 71.5% of AI Gateway tokens and 23% of spend
Open-weight models now process more than seven in ten tokens on Vercel's AI Gateway and account for nearly one in four dollars spent, according to the company's October…

Vercel Token Pass gives Pro teams $80 in model credit for $30 a month
Vercel has launched Token Pass in early access for Pro teams. For $30 per month, a team receives $80 in monthly credit for the GLM 5.3 and Kimi K3 models, according to…

Goodfire's probe-based cyber monitors beat an LLM judge at 50x lower cost
Goodfire Research says it has built cybersecurity monitors for Kimi K3 and GLM 5.3 that outperform an LLM judge at about 50x lower cost, even when run live on every…

Together Link Puts Open Models Inside Claude Code, Codex and Other Coding Agents
Together AI has released Together Link, a tool that connects coding agents teams already use to open models hosted on its platform. According to the company, it can cut…

Kimi K3 under fire: 2.8T-parameter open model faces distillation claims
Moonshot AI's Kimi K3, a 2.8 trillion-parameter open-weight model released on July 16, 2026, is now at the center of a dispute over whether it was built by distilling…

Harness-Zero study trains an agent harness into a small model
Researchers at Peking University trained the behavior of a specialized agent harness into a small model, then removed the harness, and the model outperformed the…

Fireworks' Ember-1 reasoning model arrives on Vercel's AI Gateway
Vercel has added Ember-1 from Fireworks to its AI Gateway. Ember-1 is a research preview reasoning model built on Kimi K3 for coding and agentic workflows, and Fireworks…