Summary
Command Code currently rejects adjustable reasoning effort for both Muse Spark 1.2 routes, even though the live Meta endpoints exposed through OpenRouter advertise reasoning_effort support for those same model IDs.
Affected models:
meta/muse-spark-1.2meta/muse-spark-1.2-contributor
This prevents reproducible cost, latency, and benchmark comparisons. In particular, it is not possible to reproduce evaluations such as Artificial Analysis's explicitly labeled Muse Spark 1.2 xhigh run through Command Code.
Expected Behavior
Command Code should accept the reasoning efforts supported by the serving endpoint:
minimallowmediumhighxhigh
For example, this should run with explicitly selected effort:
cmd -p "Return exactly OK" \
--model meta/muse-spark-1.2-contributor \
--effort xhigh
If Command Code intentionally restricts effort selection for these routes, the reason and effective default should be documented.
Actual Behavior
Every tested effort is rejected before inference:
Muse Spark 1.2 Contributor has no adjustable reasoning effort.
The same behavior occurs with minimal, low, medium, high, and xhigh. The Command Code model catalog also displays no supported efforts for regular Muse Spark 1.2 or the Contributor route.
Steps to reproduce the issue
-
Authenticate to Command Code.
-
Run:
cmd -p "Return exactly OK" \
--model meta/muse-spark-1.2-contributor \
--effort medium
-
Observe:
Muse Spark 1.2 Contributor has no adjustable reasoning effort.
-
Repeat with minimal, low, high, or xhigh; all are rejected.
-
Optionally repeat with meta/muse-spark-1.2.
Command Code Version
1.32.1
Operating System
Linux
Terminal/IDE
pi / tmux
Shell
bash
Session file (optional)
Not included; this reproduces with a one-shot headless command.
Fix prompt (optional)
Audit the model-capability metadata and request mapping for meta/muse-spark-1.2 and meta/muse-spark-1.2-contributor. The live provider endpoints advertise reasoning and reasoning_effort, with supported efforts minimal, low, medium, high, and xhigh, defaulting to medium. Expose those values through --effort and /effort, pass the selected value to the provider, add headless coverage for all accepted values, and document the effective default. If the Command Code backend intentionally cannot pass this parameter, document the restriction instead.
Additional context
OpenRouter's live catalog reports both routes as mandatory-reasoning models with configurable effort and a default of medium:
The endpoint metadata lists both reasoning and reasoning_effort under supported_parameters. This suggests a Command Code model-metadata or provider-mapping mismatch rather than an underlying Muse capability limitation.
No credentials or private session data are needed to reproduce the rejection.
Summary
Command Code currently rejects adjustable reasoning effort for both Muse Spark 1.2 routes, even though the live Meta endpoints exposed through OpenRouter advertise
reasoning_effortsupport for those same model IDs.Affected models:
meta/muse-spark-1.2meta/muse-spark-1.2-contributorThis prevents reproducible cost, latency, and benchmark comparisons. In particular, it is not possible to reproduce evaluations such as Artificial Analysis's explicitly labeled Muse Spark 1.2
xhighrun through Command Code.Expected Behavior
Command Code should accept the reasoning efforts supported by the serving endpoint:
minimallowmediumhighxhighFor example, this should run with explicitly selected effort:
cmd -p "Return exactly OK" \ --model meta/muse-spark-1.2-contributor \ --effort xhighIf Command Code intentionally restricts effort selection for these routes, the reason and effective default should be documented.
Actual Behavior
Every tested effort is rejected before inference:
The same behavior occurs with
minimal,low,medium,high, andxhigh. The Command Code model catalog also displays no supported efforts for regular Muse Spark 1.2 or the Contributor route.Steps to reproduce the issue
Authenticate to Command Code.
Run:
cmd -p "Return exactly OK" \ --model meta/muse-spark-1.2-contributor \ --effort mediumObserve:
Repeat with
minimal,low,high, orxhigh; all are rejected.Optionally repeat with
meta/muse-spark-1.2.Command Code Version
1.32.1
Operating System
Linux
Terminal/IDE
pi / tmux
Shell
bash
Session file (optional)
Not included; this reproduces with a one-shot headless command.
Fix prompt (optional)
Audit the model-capability metadata and request mapping for
meta/muse-spark-1.2andmeta/muse-spark-1.2-contributor. The live provider endpoints advertisereasoningandreasoning_effort, with supported effortsminimal,low,medium,high, andxhigh, defaulting tomedium. Expose those values through--effortand/effort, pass the selected value to the provider, add headless coverage for all accepted values, and document the effective default. If the Command Code backend intentionally cannot pass this parameter, document the restriction instead.Additional context
OpenRouter's live catalog reports both routes as mandatory-reasoning models with configurable effort and a default of
medium:The endpoint metadata lists both
reasoningandreasoning_effortundersupported_parameters. This suggests a Command Code model-metadata or provider-mapping mismatch rather than an underlying Muse capability limitation.No credentials or private session data are needed to reproduce the rejection.