[Windows][VS Code extension] bundled Codex 0.147.0-alpha.6.5 rejects gpt-5.6-luna for spawn_agent while standalone 0.147.0 succeeds

What issue are you seeing?

gpt-5.6-luna cannot be used as a subagent when running the Codex binary bundled with the current VS Code extension.

On the same Windows machine, with the same account, project, and ~/.codex/config.toml, standalone stable Codex CLI 0.147.0 successfully spawns gpt-5.6-luna from a gpt-5.6-sol parent.

The VS Code extension-bundled Codex binary rejects Luna with:

Unknown model gpt-5.6-luna for spawn_agent. Available models: gpt-5.6-sol, gpt-5.6-terra

Importantly, the failure also reproduces when the exact VS Code extension-bundled codex.exe is launched directly from PowerShell, without VS Code or the extension app-server. This appears to rule out a VS Code UI/app-server-specific cause and distinguishes this case from a stale older app-server process.

Environment

VS Code extension:

openai.chatgpt@26.803.61601

Bundled Codex CLI:

codex-cli 0.147.0-alpha.6.5

Standalone Codex CLI:

codex-cli 0.147.0

Platform:

Windows 11 x64
VS Code

Parent model:

gpt-5.6-sol

Requested subagent model:

gpt-5.6-luna

Relevant configuration

model = "gpt-5.6-sol"
model_reasoning_effort = "xhigh"

[agents]
enabled = true
default_subagent_model = "gpt-5.6-luna"
default_subagent_reasoning_effort = "high"
max_concurrent_threads_per_session = 6

Reproduction

1. Standalone stable CLI works

codex --version
# codex-cli 0.147.0

codex -m gpt-5.6-sol

Ask the Sol parent to explicitly spawn exactly one gpt-5.6-luna subagent, with no fallback to Sol or Terra.

Result: Luna subagent starts successfully.

2. VS Code extension fails

The installed extension is:

code --list-extensions --show-versions | Select-String -Pattern "openai|chatgpt|codex"

Result:

openai.chatgpt@26.803.61601

Process inspection shows the extension actually starts its bundled binary:

...\.vscode\extensions\openai.chatgpt-26.803.61601-win32-x64\bin\windows-x86_64\codex.exe
-c features.code_mode_host=true app-server --analytics-default-enabled

The bundled executable reports:

codex-cli 0.147.0-alpha.6.5

Attempting the same Luna spawn through the VS Code extension fails with:

Unknown model gpt-5.6-luna for spawn_agent. Available models: gpt-5.6-sol, gpt-5.6-terra

3. Launch the extension-bundled Codex binary directly

To determine whether this is specific to the VS Code app-server path, launch the exact bundled executable directly:

$vsCodeCodex = "C:\Users\<USER>\.vscode\extensions\openai.chatgpt-26.803.61601-win32-x64\bin\windows-x86_64\codex.exe"

& $vsCodeCodex --version
# codex-cli 0.147.0-alpha.6.5

& $vsCodeCodex -m gpt-5.6-sol

Then request:

Spawn exactly one subagent using gpt-5.6-luna.
Do not fall back to Sol or Terra.
If spawning fails, return the spawn_agent error unchanged.

Result:

Unknown model gpt-5.6-luna for spawn_agent. Available models: gpt-5.6-sol, gpt-5.6-terra

Therefore the failure reproduces in the extension-bundled binary itself, even when it is run directly without VS Code.

A/B result

| Runtime | Luna spawn |
|---|---|
| Standalone codex-cli 0.147.0 | ✅ Works |
| VS Code extension bundled 0.147.0-alpha.6.5 through app-server | ❌ Fails |
| VS Code extension bundled 0.147.0-alpha.6.5 launched directly | ❌ Fails |

All tests use the same Windows machine, account, configuration, and project environment.

Expected behavior

The Codex binary bundled with the VS Code extension should allow a gpt-5.6-sol parent to spawn gpt-5.6-luna, consistent with standalone stable Codex CLI 0.147.0 and the leaf-model support added for Multi-Agent V2.

Actual behavior

The extension-bundled 0.147.0-alpha.6.5 binary exposes only:

gpt-5.6-sol
gpt-5.6-terra

as valid spawn_agent models and rejects gpt-5.6-luna.

Related issues / changes

• #34964 — spawn_agent does not expose gpt-5.6-luna
• #35097 — Luna marked as MultiAgent V1 caused V2 spawn_agent rejection
• #37536 — stale older app-server can cause pre-0.147 Luna filtering
• #36892 — support leaf models in Multi-Agent V2

This appears different from #37536 because:

1. The standalone stable 0.147.0 CLI works.
2. The VS Code extension is actually running its bundled 0.147.0-alpha.6.5 binary.
3. The extension-bundled binary fails even when launched directly, without VS Code or app-server reuse.
4. No stale older app-server is required to reproduce the issue.

It appears that the Codex build currently bundled with VS Code extension 26.803.61601 does not contain, or does not activate, the same Luna/leaf-model spawn support as stable Codex CLI 0.147.0.