Google introduced Gemini 4 Argon, a new frontier model with a 1 million token output limit, strong coding and cybersecurity benchmark results, and a phased rollout starting with trusted cyber defenders.
xAI released Grok 4.7, a coding- and knowledge-work-focused model that works longer on hard tasks and checks its own results more carefully, matching Grok 4.6's pricing while adding what xAI calls an entirely new safeguard stack with better jailbreak resistance.
OpenAI and AWS have brought the GPT-5.6 model family to Kiro, AWS's spec-driven AI coding agent, with OpenAI citing an 82% cost reduction on Terminal-Bench 2.1 tasks.
Replit is powering its Free Mode with GPT-5.6 Luna, giving free users fast, no-usage-cost AI coding assistance that escalates to GPT-5.6 Sol for harder reasoning tasks while preserving full project context.
xAI's coding model Grok 4.5 has been added to GitHub Copilot's model picker. Available in VS Code, Copilot CLI, and cloud agents, it's built for fast, agentic coding tasks that dispatch multiple tools in parallel.
xAI has fully open-sourced its coding agent "Grok Build." The move comes in response to a recently discovered issue involving unauthorized repository uploads, and usage limits have also been removed.