GPT-5.6 is now the default model in ChatGPT. Agents are the job.

1. GPT-5.6 is what most people will now use

On August 6, OpenAI moved ChatGPT onto the GPT-5.6 family. Free and Go users get Luna as the default, unlimited text chats, and a Think button for harder questions. Plus and Pro get Sol with a slider for how much reasoning to spend. That replaces GPT-5.5 Instant in chat. OpenAI’s post and the help article are the source of truth.

If you write “used ChatGPT” on a resume, name the mode. Luna vs Sol vs Think is the difference between a first draft and something you would send.

2. Claude’s current models are built to run, not just reply

Opus 5 (July 24) is the Max default for long coding and knowledge-work agents. Sonnet 5 is the Free and Pro default. Anthropic describes it as a model that plans, uses a browser and a terminal, and keeps going. That is a different skill than writing a clever prompt.

The brief that works now: goal, tools, stop condition, what a human must check. Paste one from the agentic library.

3. If you live in Microsoft 365, agents are a product, not a demo

Microsoft’s Copilot agents overview (updated August 5) is the current map: declarative agents in Agent Builder vs custom-engine agents when you need your own orchestrator. New Agent Builder agents get core capabilities on by default in a late-August rollout. That is how these things will show up in your company’s agent store.

Pick one recurring task you already own. Write the instructions, the knowledge sources, and the approval step. That artifact is more useful in a review than another certificate.

Gemini, for the record

The current flagship to check is Gemini 3.1 Pro, with Deep Think for hard science and engineering work. Use it when the job is Google Workspace plus long context, not because the name is newer.