For a one-off build, code it with Claude Opus 5.

Why Claude Opus 5

The reasons it wins for this job.

Anthropic positions Opus 5 "for complex agentic coding and enterprise work" and tells developers to start with it for most workloads. It is the default model in Claude Code on the Max plan.
On Terminal-Bench 4.0, which measures real multi-step work in a terminal, Opus 5 scores 52.3%. Only Claude Fable 5.1 (55.8%) and GPT-6 Astra (57.9%) score higher, and both cost twice as much per token.
A 1M-token context window means it can hold a mid-sized codebase in view, so it edits the right files instead of guessing at them.
For a single build, speed matters more than the last few points of benchmark. Opus 5 is noticeably faster than Fable 5.1, which Anthropic describes as its slowest model.

How to use it

4 steps to a first result.

1Install Claude CodeRun npm install -g @anthropic-ai/claude-code and sign in with your Claude account. Opus 5 is selectable with /model.
2Open the repo, not a chat windowStart claude inside the project folder so it can read files, run the dev server and run the tests itself.
3Give it the finished pictureDescribe the outcome, the constraints and how you will check it. A task with a test or a screenshot to match finishes faster than a vague one.
4Review the diff before you shipAsk it to run the build and tests, then read the diff. Commit when you would have approved it from a colleague.

Prompt for Claude Code

Start from this.

Edit the parts in capitals, then run it.

Rebuild PAGE OR FEATURE in this repo.

What done looks like:
- DESCRIBE THE RESULT A USER SEES
- It matches the existing design system (reuse our components, no new CSS framework)
- `npm run build` and `npm test` pass

Constraints:
- Do not change anything outside FOLDER
- No new dependencies without asking me first

Start by reading the relevant files and telling me your plan in five bullets. Wait for my OK, then build it, run the build and tests, and show me the diff.

Nothing is sent anywhere. It copies to your clipboard.

Alternatives that also work

If Claude Opus 5 is not an option.

GPT-6 AstraOfficial page →

OpenAI · closed · cost: high

Pick it when your team already lives in ChatGPT or Codex. It tops Terminal-Bench 4.0 at 57.9% and matches Fable 5.1 on the Artificial Analysis Intelligence Index.

Claude Fable 5.1Official page →

Anthropic · closed · cost: high

Pick it when the job is long and hard, such as a migration across hundreds of files, and you can wait. It is Anthropic's most capable model and its slowest.

GLM-5.3Hugging Face →

Z.ai · open weights · cost: free weights; you pay for the hardware · 838K HF downloads/mo

Pick it when the code cannot leave your own infrastructure. It is the strongest open-weight coding model on Terminal-Bench 2.1 (88.2) by its own model card.

Watch out for

Benchmarks have moved. SWE-bench Verified is close to saturated, so labs now publish Terminal-Bench, DeepSWE and CursorBench. Compare models on the same benchmark, never across two.
A one-off script often becomes a recurring job. When it does, read the recurring page before you wire it into anything.

Doing this again and again? For coding that runs every day, use Claude Sonnet 5. Automated code review, test generation and small fixes run hundreds of times. Sonnet 5 gives near-Opus coding quality at a medium cost tier with the same 1M context, which is the trade a pipeline needs.

Sources

Checked 17 September 2026. Models change monthly; we re-check this page when they do.

Picking the model is the easy part.

Wiring it into a workflow that runs every week, with evals, fallbacks and a cost you can predict, is the work. Fifteen minutes, no deck.

Book fifteen minutes →