What is a frontier model? The most capable one now

Nov 8, 2026

A frontier model is one of the most capable models right now, and the label moves. How Claude Code, Codex, Cursor and Grok name the model they point you at.

A frontier model is one of the most capable models you can use right now. The label moves when a newer model takes that place.

It is not a setting. Cursor’s docs and the Grok 4.7 card use the phrase. The Claude Code and Codex pages checked for this entry do not.

The AI coding landscape is the map of who ships these models. This entry is the label, and which alias each tool tells you to pick.

How a frontier model works

A vendor uses the phrase for the top of the current lineup. The previous model stays available. You still choose per task. A long migration and a one-line rename do not need the same model.

Exam scores stopped separating the top models on some tests. How AI models are measured is that argument, with the charts.

Frontier models in each tool

	Claude Code	Codex	Cursor	Grok
The phrase	Not on the model config page	Not on the models page	“Frontier models” from OpenAI, Anthropic, Google, SpaceXSI, and more	The Grok 4.7 card says “SpaceXSI’s frontier model”
What to run for the hardest work	`fable` for the hardest and longest tasks. `opus` for complex reasoning	Astra for the most demanding workflows	The model you pick. The agent page says Cursor tunes tools for every frontier model	Grok 4.7, which the models index calls the most capable model they have built

Claude Code

The model config page does not say “frontier model”. Aliases pick a class. fable uses the Fable model for your hardest and longest-running tasks. opus uses the latest Opus for complex reasoning. sonnet is for daily coding. haiku is for simple tasks. best uses the model fable resolves to where Fable is available to you, otherwise the same model as opus.

To select Fable 5.1, the page says run /model fable, or launch with claude --model fable. See Model configuration.

OpenAI Codex

The models page does not say “frontier model”. The recommendation there, in November 2026: Astra suits the most demanding workflows. GPT-6.1 Sol offers near-Astra performance at a lower cost than Astra. GPT-6.1 Terra is the visual one, covered under multimodal model. Luna suits focused, repeatable tasks.

In an interactive session, /model switches the model or the reasoning effort. codex --model and codex -m set it at launch. See the Codex models page.

Cursor

The models and pricing page opens with “Cursor supports frontier models from OpenAI, Anthropic, Google, SpaceXSI, and more.” The agent page says Cursor tunes instructions and tools for every frontier model.

Neither sentence names a single winner. You still pick the model for the task. See Models and pricing and the agent overview.

Grok

The Grok 4.7 page calls it “SpaceXSI’s frontier model for coding, agentic tasks, and knowledge work.” The models index says that for everything else, including code, you use Grok 4.7, and that it is the most capable model they have built.

/model in Grok Build switches models. See Grok 4.7 and the models index.

Details checked against each vendor’s documentation in November 2026.

A real example

A long task is the case the Claude Code alias table points at Fable:

claude --model fable

That command selects Fable 5.1, per the model config page. It does not make the phrase “frontier model” appear in Claude Code’s docs.

For a repeated coding pass where Codex’s own page says you do not need Astra, the same page points at GPT-6.1 Sol. Switch with /model in the session.

When to use a frontier model

Use the hardest alias when the task is long, the design is open, or a wrong guess is expensive to unwind. The Claude Code table says fable and opus for that kind of work. Codex says Astra.

Use the smaller alias for a focused, repeatable edit. Claude Code says haiku for simple tasks. Codex says Luna for focused, repeatable tasks. Cursor leaves the pick to you.

Pitfalls

The label moves. A model that was the top of the lineup last season is a normal choice now. Read the model page again before you quote a name.

The phrase is not a shared mode. Astra and Fable are the hardest option on those pages. The pages do not call them frontier models.

A high score on a saturated exam is not a reason to pick a model. The benchmarks post is where that chart lives. Which AI model is best for coding? is the hands-on comparison on this site.

Prices move, and they are not in this entry. The model pages above are the place to read them.

Coding agent: A coding agent uses a model and tools to read a repo, edit files and run commands in a loop. How Claude Code, Codex, Cursor and Grok Build each do that job.
Evals: An eval is a fixed prompt, a real agent run, and a check you can repeat. How Claude Code, Codex, Cursor and the Grok API score that run.
Multimodal model: A multimodal model takes more than text, such as a screenshot, in the same request as the prompt. How Claude Code, Codex, Cursor and Grok accept an image.
Reasoning effort: Reasoning effort sets how long a model thinks before it answers. The levels in Claude Code and Codex, what Cursor's plans fix, and how Grok sets effort.

All AI glossary entries →

Follow @flaviocopes

Want me to talk about your product? You can sponsor this site.