Codex + GPT-5.6 Luna
GPT-5.6 Luna is currently selectable with Codex on AgentSky. Codex controls the work loop and tools; the model supplies reasoning and generation. Start a real task below with both choices already selected.
How the combination works
Codex runs the work. GPT-5.6 Luna powers each turn.
Codex owns the coding-agent loop: it explores the repository, changes files, runs commands, and returns evidence from the cloud workspace.
AgentSky supplies the persistent cloud computer, task state, usage records, and enabled tools or connectors around that combination.
Codex
Owns the agent loop, repository or computer operations, tool calls, and how results are returned.
GPT-5.6 Luna
OpenAI's model supplies the reasoning and generated output within Codex's supported interface.
AgentSky
Keeps the selected pair, workspace, task state, access configuration, and itemized usage together.
Current GPT-5.6 Luna facts
What the model is
The cost-optimised GPT-5.6 tier, built for high-volume workloads.
Context window
1.1M tokens in the current AgentSky catalog.
Accepted input
Text, Image in the current AgentSky catalog.
Model token price
$0.2 / $1.20 per 1M input / output tokens on the displayed AgentSky plan. Runtime and paid tool usage are separate.
Evaluate it on real work
Start with one bounded task and evidence you can inspect.
Provide the repository, base revision, requirement, and a concrete success check. State whether the task is review-only or may change the branch.
Open the preselected Composer
The Composer above starts with Codex and GPT-5.6 Luna. Add the repository or project and only the tools the task needs.
Adapt this first task
Trace the affected callers for this requirement, implement the smallest complete change, run the relevant checks, and report the exact files changed plus any remaining risk.
Verify the outcome
Review the diff against the original requirement and rerun the checks that observe the user-visible behavior. Treat a plausible answer as a proposal until the evidence supports it.
Fit and limits
When to choose Codex + GPT-5.6 Luna
Software engineering tasks that benefit from Codex's repository navigation, command execution, and iterative implementation loop.
This page establishes
- The pair is selectable in the current AgentSky compatibility matrix
- The launch carries both selections into the shared Composer
- Current catalog facts are shown when the gateway publishes them
You still need to measure
- Quality on your own repository or workflow
- End-to-end task time, recovery effort, and human review
- Total task cost, which also depends on runtime, tools, connectors, and token usage
Common questions
Related resources