OpenAI released three GPT-5.6 variants: Sol ($5/$30), Terra ($2/$12), and Luna ($0.20/$1.20) with cached-input and batch discounts. Developers now have sub-$1M-token entry pricing on the flagship model, forcing cost comparison across Claude Sonnet 5 ($2/$10) and shifting production routing calculus.

Z.ai released an open-weight model competitive with frontier Anthropic and OpenAI systems for coding and cybersecurity automation. Open availability reshapes cost-per-capability for security-focused builders and signals accelerated capability parity outside the US lab duopoly.

Llama 3.1 8B Instruct is now $0.02 per million input tokens; Llama 4 Maverick flagship is $0.15. Open-weight pricing floor dropped 10x since 2024, forcing hosted-model APIs to compete on latency, throughput, and reliability rather than raw cost—structural shift for edge and embedded deployments.

agent-device launched as a cross-platform runtime (iOS, Android, macOS, Linux, web) compatible with Claude Code, Codex, Cursor, and MCP-compatible agents for automated app testing. Developers at Expensify and Shopify now have standardized agent-device automation—enabling reliable CI/CD integration for mobile AI workflows.