Kimi releases K3.1 Pro, a 3T-class open-weight model with a 1-million-token context window, on Kimi.ai, Kimi Work, Kimi Code and the Kimi API. The company says it scores 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra at 58.2%. A Highspeed variant runs 2-3× faster. Weights are due by Nov 4.
Moonshot releases Kimi K3.1 Pro, an upgrade to its open-weight Kimi K3, with full weights due by 4 November. Moonshot says the model scored 65.7% on Terminal-Bench 4.0, just behind Claude Opus 5.5 at 66.4%, and that it stripped cyberattack training data before the open release.
Anthropic donates Agent Skills, its open standard for giving agents task-specific skill files, to the Agentic AI Foundation under the Linux Foundation. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, and GitHub Copilot, have adopted it, with 1 million skills published. David Soria Parra and Mahesh Murag are lead maintainers.
Perplexity Research and NVIDIA propose JOLT, a self-distillation method that trains one model as both teacher and student. They claim JOLT+ beats GRPO by 4.2 percentage points on MATH and 9.9 points on the AppWorld state-test pass rate. On MATH, JOLT uses 6.5 times fewer processed training tokens at matched accuracy.
Ant Group and the vLLM community open-source Humming, a kernel library for quantized inference that supports NVIDIA GPUs from Turing onward and T-Head's Zhenwu 810E. Kernels compile just in time from shared CUDA components. The groups claim switching DeepSeek-V4.1-Flash MoE execution to Humming's W4A8 path on eight H20 GPUs raises throughput 14% to 43%.
Base Labs releases mammoth-1 and elephant-1, two decision models built by fine-tuning Qwen3.5-27B and Qwen3.5-9B on 6.21 million examples, with weights published. Base Labs says mammoth-1 ranks first of 208 models on the Jev Decision Index 0.4 at 71.24, 1.8 points ahead of Perplexity Decider v1.2, and that elephant-1 scores 62.65, best under 15B parameters.
Peter Steinberger, OpenClaw's creator and an OpenAI staff member, says he spent roughly $10 million on tokens last month, money he says comes from investors. He said it at Cloudflare Connect 2026 in San Francisco. Cloudflare CTO Dane Knecht noted few developers have such a budget.
LiteLLM says attackers compromised its release pipeline and published litellm 1.107.0 to PyPI and Docker Hub with credential-stealing malware from 11:39 to 13:23 UTC on Oct 21. The packages are pulled. The company advises pinning to 1.106.2 and says enterprise registry customers were not affected.
OpenAI says its Patch the Planet coalition coordinated fixes for critical PHP vulnerabilities that Calif found with GPT-5.6 Sol and that Wiz independently reported. PHP maintainers led the patches. OpenAI committed $1 billion in subsidized Daybreak access for Frontline Defenders.
EAGLE-4 publishes inference code and eagle-harness, which adds its speculative decoding to SGLang, vLLM and TokenSpeed. The EAGLE-4 Team claims up to 3.42x faster generation on Qwen3.8-27B with four B300 GPUs. The source-available license permits commercial use below $20 million in aggregate revenue.