AI Product Updates: New Features and Releases

Odyssey-3 generates interactive environments in real time where objects move and react to user actions, now in research preview and being tested for robotics and self-driving training.

Crossmint's Agent Commerce Toolkit lets agents save any card and check out at any online store through a single integration, live today.

Dreamwork's job platform released an MCP server that lets Claude search jobs, explore roles and apply to companies without leaving the chat.

Google released the Gemini agent, which is designed to take an objective rather than a list of instructions and plan and execute on its own.

Compound V2 is out, offering collaborative AI analysts, a cloud Office suite, and large-scale tasks that run hundreds of subagents in parallel.

StepFun's Step 5 Preview model is now available on OpenRouter, Kilo Code, Cline, Hermes Agent, and OpenCode, with one user building an app from a single prompt in Hermes Agent for about $1.

AI evaluation platform Arena closed a $200M Series B at a $3.1B valuation and introduced the Arena Alignment Index, extending its evaluations from capability to agent safety and faithfulness to human intent.

Goodfire's new monitors inspect the model's internal state while an agent works and only call in extra checks when something looks suspicious, instead of paying a second AI to read everything the agent does.

Dreamwork's MCP service gives agents access to 2 million live roles, tailors resumes and submits applications directly to employers, and can be added with a single npx command.

Hedera's Agent Accounts let agents like Claude Code and Codex hold funds, pay, settle and trade autonomously under user-set spending limits, time windows, recipient allowlists and 2FA for large transactions, with a Wallet SDK and API.

Built from over 90K real-world agent sessions, the benchmark scores unauthorized actions, false attribution and deceptive completion, with GPT-6.1-Sol leading at 87.9.

Interface is a handheld device pitched as the primary hardware for the agent era, letting you control all your agents from one hand for $99, shipping in January.