Skip to main content
Managed Research supports two public harnesses:
  • codex
  • opencode_sdk
Model compatibility is harness-specific. Backend preflight remains authoritative.

​
Codex harness

Model IDReasoning effort supportDefault
gpt-5.3-codexmediummedium
gpt-5.3-codex-sparkmediummedium
gpt-5.4medium, highmedium
gpt-5.4-minilow, medium, highmedium
gpt-5.4-nanolowlow
gpt-oss-120bhighhigh
Example: 
run = client.runs.start(
    "Review the repo and leave evidence for the smallest safe fix.",
    host_kind="daytona",
    work_mode="directed_effort",
    providers=[{"provider": "openrouter"}],
    agent_harness="codex",
    agent_model="gpt-5.4",
    agent_model_params={"reasoning_effort": "high"},
)

​
OpenCode SDK harness

Model IDReasoning effort support
anthropic/claude-sonnet-4-6None in the catalog
anthropic/claude-haiku-4-5-20251001None in the catalog
x-ai/grok-4.1-fastNone in the catalog
moonshotai/kimi-k2.6None in the catalog
Example: 
run = client.runs.start(
    "Review the repo and propose the smallest high-impact fix.",
    host_kind="daytona",
    work_mode="directed_effort",
    providers=[{"provider": "openrouter"}],
    agent_harness="opencode_sdk",
    agent_model="anthropic/claude-sonnet-4-6",
)

​
Source of truth

The backend-owned Managed Research model catalog is the source of truth. The generated SDK enum in managed-research mirrors public model IDs, but backend preflight is the final authority for launch availability.