GenAI Field GuideGetting started
Choosing a model
Pick the model in three steps: what your platform offers, what shape the task is, and how hard the thinking is. The newest model is rarely the deciding factor for office work; the document you loaded and the check you ran matter more.
Step 1: what is available to you
For official work the only models that exist are the ones accredited on your platform. A newer version may be public before it reaches you; that is normal. Use what is on the portal, and follow your command's guidance on anything else.
Step 2: the shape of the task
| Task shape | Reach for | Why |
|---|---|---|
| Draft, rewrite, iterate text | The text-drafting workhorse (ChatGPT Mil today, or Gemini) | Mature projects and custom-helper workflow |
| Long documents, decks, scans, video | The long-context, multimodal model (Gemini today) | Takes the whole file without you splitting it; still say which section your question is about |
| Hard analysis, tricky logic, code | The strongest model your platform offers, with thinking turned up | This is the one place a generation gap shows |
| Bulk, simple, repetitive | The fast tier (Flash, mini) | Quicker, and the quality gap is small on easy work |
| Research with many searches, "keep going until done" | Agent-style tooling (Grok today), then verify hard | Built to take many steps; more places to go wrong |
| A reusable helper for your shop | Whatever builder your platform has (Gemini Agent Designer today) | The builder is a harness feature, not a model feature |
Step 3: how hard is the thinking
Many products offer a reasoning or thinking setting, sometimes in levels; some decide for you. Where you have the control, turn it up for multi-step logic, code, and anything where a wrong answer costs you. Leave it off for rewriting an email; it only makes the reply slower. See Thinking modes.
The two questions everyone asks
"Why would I pick Grok over ChatGPT?" Not because one is smarter. Because the job fits: multi-step research with lots of searching leans on Grok's agent tooling; a memo leans on ChatGPT's drafting workflow; a 200-page PDF leans on Gemini's context. The deciding factor is usually a harness feature, and every answer ends with "then verify."
"Why use the version on the portal when a newer one is in the news?" Because the portal's version is the one accredited for your work, and for drafting, summarizing, and everyday questions the difference between a 5.x and the next generation is small. Generations matter most on hard reasoning. If you have a genuinely hard problem and the newer model is available and authorized, use it with thinking on. Otherwise, load the source and check the output; that beats a model upgrade every time.