LLM

Claude Opus 4.6 achieved a 50%-time-horizon of approximately 14.5 hours on METR's software task benchmark — beating all predictions and suggesting a doubling time of under 3 months for AI task capabilities.

A new open-source project called ntransformer enables running the 140GB Llama 3.1 70B model on a single consumer RTX 3090 by streaming weights directly from NVMe storage to GPU, completely bypassing CPU RAM.

Andrej Karpathy coined a new term for OpenClaw-like AI agent systems: "Claws." Just as LLM agents were a new layer on top of LLMs, Claws provide orchestration, scheduling, persistent context, and tool calls on top of LLM agents.

Claude Code has grown to over $2.5 billion in annualized run-rate revenue as of February 2026, more than doubling since its first six months. The AI coding agent now accounts for over half of all enterprise spending on Anthropic and users average 20 hours per week with the product.

xAI released Grok 4.20 as a public beta on February 17, introducing a continuous post-deployment learning architecture that updates the model weekly from user feedback. The release also adds a four-agent collaboration system and medical document analysis via photo upload.

Anthropic released Claude Code Security on February 20, a research preview that uses Claude Opus 4.6 to reason about codebases like a human security researcher, finding over 500 previously undetected vulnerabilities in production open-source projects. The launch sent cybersecurity stocks tumbling up to 9%.

GitHub announced that Anthropic's Claude Sonnet 4.6 is now generally available in GitHub Copilot. Early testing shows excellent performance for agentic coding and search operations in VS Code and Copilot CLI.

Google DeepMind announced Gemini 3.1 Pro, featuring major improvements to overall model intelligence for tackling tougher problems. Rolling out to Google AI Pro and Ultra subscribers in the Gemini app and NotebookLM, with API preview in Google AI Studio.

Alibaba launched Qwen 3.5 on February 16 under Apache 2.0, featuring 397B parameters with a sparse MoE architecture (17B active), 256K context, and native multimodal capabilities matching leading US proprietary models on key benchmarks.

Anthropic launched Claude Sonnet 4.6 on February 17, offering major upgrades in coding, computer use, and agent planning—now the default model for Free and Pro users at the same $3/$15 per million tokens pricing.

A high-signal LocalLLaMA thread points to llama.cpp Discussion #19759, where maintainers say the ggml team is joining Hugging Face while continuing full-time support for ggml and llama.cpp.

Anthropic has released Claude Code Security in limited research preview, targeting vulnerability discovery and patch suggestion workflows while keeping human approval at the center.