본문 바로가기
News · 2026.10.02

Compare an always-on agent that picks its own tools without model selection against the new half-price models, and rework the model-and-cost mix in your workflow

AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know — Matt Wolfe · Read the original ↗

This page is a translation of the Korean original. Read the Korean original →

What you get

  • Gauge how an agent that alerts you first would work when your mail, Slack and calendar are scattered
  • Get a basis, in cost per task, for judging whether a cheaper model can deliver the same quality
  • Build the habit of checking which effort setting a benchmark came from into your evaluation process

When to use it

The source cites two situations: when you want a single agent to proactively gather and surface information scattered across work tools such as email, Slack and calendar, and when you want similar quality from a cheaper model instead of a flagship model for coding and automation work.

How to do it

  1. In the ChatGPT app, open your Dot under new chat, give it a name and avatar, and activate it.
  2. Let Dot use the connectors you have already linked to ChatGPT, such as Notion, Slack, Gmail and Google Calendar, to cut onboarding time.
  3. When you hand it email, state constraints explicitly, such as no sending and drafts only.
  4. If you need a low-cost model, use GPT-6.1 Sol in work mode and codex, but plan on the assumption that it is not yet available in regular chat.
  5. Do not run Sonnet 5.5 at max effort; if you need that level, use Opus.

What it means for my work

If you run agents, the questions are what value an always-on, alert-style agent adds in an environment where connectors are already attached, and whether constraints such as no sending actually hold when given as instructions. If you design evaluations, add a check that starts with which effort setting a vendor's published score was measured at, and whether the cost per task at that setting is no higher than the higher-tier model's.

Background — what the original covers

Matt Wolfe walks through a week of announcements centered on OpenAI DevDay 2026, which he attended in person. Alongside new features such as the always-on agent Dots, the low-cost GPT-6.1 Sol and the Decisions API, he covers Anthropic's Sonnet 5.5, Google's Gemini 4 Argon, Ideogram 4.5, voice models and the White House agreement. He also points out where the benchmark settings in the announcement materials diverge from actual pricing.

#Tools

← All practitioner tips