AI & Machine Learning
342 articles RSS
vLLM 0.31 Adds Draft-Model Speculative Decoding to Model Runner V2, a Preload CLI for Fast Restarts, and Gates Per-Request Multimodal Kwargs
vLLM 0.31.0 brings draft-model speculative decoding to Model Runner V2, adds a vllm preload CLI and a max-num-active-seqs cap, and removes tokenizer_mode="slow".
GitHub Copilot Adds Dynamic Workflows, Code-Defined Multi-Agent Orchestration, to Its CLI, App and SDK
GitHub put dynamic workflows in public preview across Copilot CLI, the Copilot app and the Copilot SDK, letting developers define in code when agents run and how results are used.
Anthropic Says Open-Weight GLM-5.3 Builds Working Exploits at Near Claude Mythos Preview Rates, With Safeguards Bypassed 64% to 100% of the Time
Anthropic reports Z.ai's open-weight GLM-5.3 developed end-to-end exploits in 50 of 410 ExploitBench attempts versus 56 for Claude Mythos Preview; NIST's CAISI calls it the most cyber-capable open-weight model.
GitHub Copilot CLI and App Add Computer Use in Public Preview, Letting the Agent Operate macOS and Windows Desktop Apps
GitHub put computer use into public preview in Copilot CLI and the Copilot app on macOS and Windows; it is off by default and asks approval before controlling an app.
Ataraxos AI Beats Stratego's Most Decorated Player 15 Wins to 1 in Nature Paper, on a Training Run Costing Under $8,000
Researchers from CMU, NYU, Stanford and MIT report in Nature that Ataraxos beat Pim Niemeijer with 15 wins, 1 loss and 4 draws, using far less compute than DeepMind's DeepNash.
CoreWeave Launches Forge, Brings Vera Rubin NVL72 Online and Says It Will Offer Vera CPU Racks for Agent Sandboxes
At its Fully Connected event, CoreWeave announced Vera Rubin NVL72 availability, plans to offer NVIDIA Vera CPUs, and the Forge platform for training and agent improvement.
Anthropic Adds Mods to Claude Code, TypeScript Functions That Rewrite Prompts, Gate Tool Calls and Run Unsandboxed
Claude Code mods, announced October 1, are plugin-delivered TypeScript functions that can rewrite prompts and block tool calls, and Anthropic says they are not sandboxed.
vLLM 0.30 Adds a Persistent GPU Weight Cache for Faster Engine Restarts and Removes GPTQ Activation Ordering
vLLM 0.30.0 adds a GPU-resident weight cache for faster engine restarts and removes GPTQ g_idx; llm-d v0.10.0 then moved its base image to 0.30.0.
OpenAI Gives Codex Reusable Cloud Environments, a Voice-Directed CLI and Offline Pull Request Reviews at DevDay
OpenAI's DevDay update to Codex adds saved cloud environments that persist across devices, a CLI with voice control and an /agents view, and automatic code reviews.
Claude Code Makes Auto Mode Its Default Starting Permission Mode on Every Plan and Provider
Claude Code 2.1.284 starts interactive terminal and VS Code sessions in auto mode on every plan and provider, extending a default that began on Pro, Max and Team plans.
Google Unveils Gemini 4 Argon, Claiming Benchmark Leads Over GPT-6 Astra and Claude Opus 5.5 but Starting With a Limited Rollout
Google announced Gemini 4 Argon, which it says leads on 12 of 18 disclosed benchmarks. Access begins with trusted cyber defenders, with introductory pricing of $2 and $10 per million tokens.
OpenAI Launches $500 Pro 500 Plan With 300-Token-Per-Second Ultrafast Codex Tier and Halves the $200 Plan's Codex Allowance
At DevDay, OpenAI added a $500-a-month Pro 500 plan with an Ultrafast Codex speed tier and will cut the $200 Pro plan's Codex usage from 20x to 10x the Plus allowance from October 30.