For everyday coding, start with GPT-6.1 Sol · Medium in ChatGPT Work/Codex or Claude Sonnet 5.5 · Medium. For simple summaries and rewrites, try GPT-6 Luna · Low or Claude Haiku 5.5 · Low. Increase effort when the task needs more investigation or checking. 146
Checked 28 October 2026. These are practical starting points based on official documentation; no head-to-head test was performed for this guide.
Quick comparison by task
The GPT column refers to Work/Codex. Ordinary ChatGPT Chat uses the labels explained below. Model access varies by product, plan and workspace. 23
| Task | GPT model and effort — Work/Codex | Claude model and effort |
|---|---|---|
| Short summary, simple rewrite or straightforward translation | GPT-6 Luna · Low | Haiku 5.5 · Low |
| Client email, website copy or tone editing | GPT-6.1 Terra · Low | Sonnet 5.5 · Low–Medium |
| Small feature, focused coding or routine implementation | GPT-6.1 Sol · Medium | Sonnet 5.5 · Medium |
| Difficult bug, architecture decision or large refactor | GPT-6 Astra · High | Opus 5.5 · High |
| Comparing long documents or conflicting sources | GPT-6.1 Sol · High; try Astra if needed | Opus 5.5 · High |
| The hardest sustained, multi-step project | GPT-6 Astra · Extra High | Fable 5.1 · High; increase if useful |
The table combines the providers’ documented model roles with an editorial choice of starting effort. It does not claim that the models in a row perform equally. 1247
What should I choose in ordinary ChatGPT?
When GPT-6 is available in Chat, try Instant for quick exchanges, Medium for everyday reasoning, and High / Extra High for harder work. Pro is the option for demanding tasks where available. These Chat controls differ from the Work/Codex model picker; GPT-6.1 Sol and Terra are separate Work/Codex options. 3
Low, Medium, High or Max?
- Low / Light: a clear, limited task with supplied information.
- Medium: several steps, a working draft or everyday coding.
- High: difficult diagnosis, competing requirements or careful verification.
- Extra High / Max: unusually demanding work, after checking whether the extra time and usage improve the result.
Effort is not a shared measurement across providers or model versions. In Work/Codex, Ultra delegates work to subagents; it is different from giving one model more time at Max. 268
More effort or a stronger model?
First provide the missing files, sources and requirements. If the assistant skips relevant checks, increase effort. If it examines the right material and still struggles, compare a stronger model. For current facts, enable search and verify the sources.
For a short visible answer, specify the format separately: “Answer in three bullets.” More thinking and more words are different choices. 89
How do I change Claude’s effort?
In Claude, open the model menu beside Send and choose Effort. In Claude Code, use /model and /effort. Current Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.1 do not offer a thinking-off switch in the Claude app; lower the effort for faster routine work. 56
For Estonian writing, test the same short sample in both tools and check natural phrasing, terminology and preserved meaning. This guide does not establish an Estonian-language winner.