Re-post-trained (same 284B MoE) for agentic work, the new build scores 82.7 on Terminal Bench 2.1 (up from 61.8) and 54.4 on DeepSWE (up from 7.3), all at $0.14/M input with MIT license and 1M-token context.
Why it matters: Best value-per-intelligence option for agentic pipelines right now; open weights make self-hosted builds viable for cost-sensitive teams.
Google's Gemini 3.1 Flash Lite and 3.5 Flash are now available through Oracle AI Agent Studio, letting Fusion Applications and NetSuite customers build and run AI agents without switching vendors.
Why it matters: Enterprise teams running Oracle ERP now have native AI agent infrastructure inside their existing stack — lowers the barrier for marketing automation use cases significantly.
Effective July 30, Bedrock Agents Classic is in maintenance mode with a frozen model catalog; new accounts must start on AgentCore, and AWS is actively pushing migration for existing users.
Why it matters: Any new AWS-hosted agent build should start on AgentCore; teams with existing Bedrock Classic deployments should begin migration planning before the model catalog gap widens.
Every audio file generated by GPT-Live through ChatGPT Voice and the API now embeds an imperceptible SynthID watermark; OpenAI also opened an API so developers can build provenance verification into their own workflows.
Why it matters: AI-generated audio disclosure requirements are tightening; this gives developer teams a technical compliance layer they can embed before regulators mandate it.