SpaceXAI moved Grok 4.5 to general availability on July 8-9, pricing it at $2/$6 per million tokens and making it default in Cursor across all plans.

• Token efficiency is the real story: Grok 4.5 uses an average of 15,954 output tokens per SWE-Bench Pro task versus Claude Opus 4.8's 67,020, producing a roughly 10x cost-per-task advantage at current pricing.
• Trained with Cursor: The model was co-developed with Cursor and targets agentic coding and long-horizon tasks, arriving the same day OpenAI pushed GPT-5.6 to broad release, setting up a direct pricing comparison.

OpenAI began the public rollout of GPT-5.6 today with three models — Sol, Terra, and Luna — after the US Commerce Department completed a security review.

• Sol is the flagship: Positioned as OpenAI's most capable model to date, with described strengths in cybersecurity, biology, and agentic coding, including vulnerability identification within internal safety boundaries.
• Terra and Luna address cost tiers: Terra targets daily use at roughly half GPT-5.5's price; Luna optimizes for speed and lowest cost, giving enterprise buyers a structured replacement path across use cases.

Mistral unveiled Robostral Navigate, an 8B-parameter model that guides robots through factories and warehouses using only a single RGB camera.

• No lidar or depth sensors required: The model runs on wheeled, legged, or flying robots and was trained entirely in simulation, removing the need for costly multi-sensor hardware setups.
• Follows the Emi AI acquisition: The launch is the first product from Mistral's May acquisition of Austrian physical AI startup Emi AI, marking Mistral's formal entry into industrial automation.

Anthropic released the Claude Apps Gateway for AWS, a self-hosted control plane that centralizes access, spend caps, and policy enforcement for Claude Code and Claude Desktop deployments.

• Eliminates per-developer credential sprawl: IT teams get a single credential management point instead of provisioning individual API keys per developer laptop, addressing a key enterprise governance gap.
• Self-hosted on AWS infrastructure: The gateway runs in the customer's own cloud environment, keeping data residency and audit trails within existing enterprise security boundaries.

Anthropic's own marketing operations team disclosed that using Claude Cowork reduced reporting and campaign setup time from days to hours by automating multi-step workflows.

• Self-reported but specific: The outcome is Anthropic eating its own cooking — a marketing ops team, not an engineering team, achieving measurable cycle-time reduction on routine campaign work.
• Agentic task design matters: The gains came from Claude Cowork's autonomous multi-step execution, not from prompt-and-respond chatbot use, illustrating the gap between AI chat adoption and AI workflow adoption.

Japanese precision manufacturer Star Precision reported that company-wide AI usage quadrupled within one month of deploying Notion and Notion AI across its 550 domestic employees.

• 90% monthly active usage rate: The company achieved near-total adoption by embedding AI directly into operational workflows including a custom QA helpdesk agent and AI meeting note analysis, rather than offering AI as a standalone chat tool.
• Workflow integration drove adoption: Prior AI tools stagnated because they operated outside existing business flows; Notion's embedded approach solved the 'AI in a sidebar' problem that plagues most enterprise rollouts.