agents · review · first seen Nov 9, 10:32 UTC

crewAI makes gpt-6-luna its default model

crewAI 1.15.38 shipped gpt-6-luna as its default model everywhere, alongside spreadsheet URL reading and changes to eval. Builders who rely on an implicit model choice should check it before accepting results from an existing workflow.

L0 · primary1 source4 claims checked4 min read

What shipped

The release notes say 1.15.38 makes gpt-6-luna crewAI’s default model everywhere. This is a default change, not a statement that every existing project selects that model. If a workflow leaves model choice to crewAI, inspect that choice before comparing its output with an earlier run. The notes do not give a migration procedure or describe how an explicit model setting interacts with the new default. [1]

An agent can read spreadsheet URLs in its input, with tools over every row. That is a new input path a builder can test with a spreadsheet URL and a row-level task. The excerpt does not name supported spreadsheet services, formats, authentication methods, row limits, or tool signatures. Treat the feature as something to exercise against your own input, not as a promise about all spreadsheets. [1]

For evaluation, crewai eval adds --no-open for headless use. The release also counts an eval run that exits with an error as run_failed. These are separate changes: one addresses opening behavior in a headless setting; the other addresses how an error-exiting run is counted. The notes also mention tagging eval usage spans with a target and a per-evaluation id, but give no further instructions for using those tags. [1][2]

Why it matters if you build

The model default is the first thing to check if your agent depends on an implicit choice. Run a representative task before accepting the new default for that workflow, and consider specifying the model if comparison across runs matters to you. This is a review step, not a claim that outputs will differ: the source offers no benchmark, price, latency, or compatibility result for gpt-6-luna. It also does not say what happens to a project that already specifies a model. [1]

Spreadsheet URLs could remove a manual handoff between a sheet and an agent’s input. A useful trial is to supply a small spreadsheet URL and inspect whether the agent’s tools address each row as intended. Do not plan capacity or access controls from this note alone. It supplies neither a supported-size limit nor a cost model, and it does not establish which URL permissions or spreadsheet variants work. A trial with your own source can resolve some of those questions for your workflow. [1]

The eval changes matter when an evaluation runs without someone watching a browser window. Use --no-open for a headless run, then check the recorded status of a run that exits with an error. The release says that case is counted as run_failed. It does not establish that every failure mode is covered or say how a reporting system displays that status. Check the result your own workflow consumes. [1][2]

How to try it today

1. Check whether your crewAI workflow relies on a default model before using 1.15.38.
2. Run one representative agent input and inspect the selected model.
3. Give an agent a spreadsheet URL and inspect its tools’ work across the rows.
4. Run crewai eval --no-open in a headless setting and inspect the result.
5. Test an eval run that exits with an error and check whether your reporting records run_failed.

What we checked

Every claim below was walked down the source ladder. The rung is how close the source is to the thing itself; the tense is the source's own.

1. shipped L0 · primary

   crewAI 1.15.38 makes gpt-6-luna the default model everywhere.

   per GitHub releases · crewAIInc/crewAI (L0)
2. shipped L0 · primary

   crewAI 1.15.38 reads spreadsheet URLs in an agent’s input with tools over every row.

   per GitHub releases · crewAIInc/crewAI (L0)
3. shipped L0 · primary

   crewAI 1.15.38 adds --no-open to crewai eval for headless use.

   per GitHub releases · crewAIInc/crewAI (L0)
4. shipped L0 · primary

   crewAI 1.15.38 counts an eval run that exits with an error as run_failed.

   per GitHub releases · crewAIInc/crewAI (L0)

What is not confirmed

The source does not establish gpt-6-luna’s price, speed, quality, availability, or behavior under an explicit model setting. It does not specify spreadsheet URL formats, access requirements, tool interfaces, or row limits. For eval, it does not document all error cases or how downstream reporting displays run_failed. None of those gaps should be treated as a shipped guarantee. [1][2][3][4]

Who should act, who should wait

Act today if you rely on crewAI’s implicit model choice, feed spreadsheet URLs into agents, or run eval without an interactive window: each has a narrow change you can test in 1.15.38. Wait before a broad rollout or budgeting decision if it depends on model performance, spreadsheet scale, or a particular reporting integration. The release supports testing these features; it does not provide the missing measurements or integration details. [1][2][3][4]

Sources

• 1.15.38 — GitHub releases · crewAIInc/crewAI L0 · primary