Why is a model unavailable on Martini?
Short answer
A model can be unavailable for three reasons: provider outage (the upstream API is down), Martini disabled it temporarily (during a known issue or rollout), or your plan doesn't include it (the model card shows a lock icon and a Requires {plan} label). The error label on the model picker tells you which.
Provider outage
Every AI model on Martini runs on the upstream provider's GPU clusters — OpenAI, Google, FAL, Runway, Hailuo, and others. When a provider has a regional outage, capacity event, or rolling maintenance, the corresponding model on Martini returns Provider unavailable until the upstream service recovers. Outages are usually model-specific, not platform-wide, so a different model from a different provider continues to work.
The fix during an outage is to switch to a sibling model from the node's model picker. Hailuo and Luma Ray are good fallbacks during a Sora or Veo outage; Seedance 2 is a good fallback during a Kling outage. Status notes for major outages are posted in the help icon notifications when they occur. If you need the specific model that is down, wait — provider outages typically resolve within minutes to a few hours.
Martini-side temporary disable
Martini will disable a model temporarily if a known issue affects output quality, billing accuracy, or safety. This happens during rollout windows for new models, after a provider pushes a breaking API change, or when a bug surfaces in cost calculation. The model picker shows the model with a Temporarily unavailable label and a brief reason. These windows are usually short — hours to a day.
When this happens, the model returns to the picker once the issue is resolved. Affected projects and workflows are not deleted; they are simply paused on that node. Switch the node to a sibling model if the work is time-critical, or wait for the model to return.
Plan gating
Some models are reserved for specific plan tiers. Sora 2, Veo 3, Runway Gen-4, and a handful of premium providers require a paid plan or a higher tier than the one you currently hold. The model card shows a lock icon and a label such as Requires Pro or Requires Standard+. Trying to run on a locked model triggers the upgrade prompt with the gate named.
The gate is a plan policy, not an outage. Either upgrade to the named tier, or pick a sibling model at your current tier. The model picker on every node lists which models are unlocked for you, with locked options grayed out and labeled. See the why-upgrade-prompt-shows article for the full set of plan gates.
Examples
- Sora 2 returns Provider unavailable for an hour during a rollout window — switch to Hailuo or wait.
- Kling shows Temporarily unavailable for half a day after a breaking API change — Martini re-enables once patched.
- Runway Gen-4 shows a lock icon labeled Requires Pro — upgrade or switch to a sibling model.
- Hailuo continues to work while Sora is down — provider outages are model-specific.
- A new model in beta release appears for some users during rollout, then for everyone at GA — gate widens over time.
Edge cases
- Workspace admins can disable specific providers across the workspace; affected models appear missing rather than locked.
- Regional restrictions imposed by a provider may hide a model in some countries — Martini reflects the provider policy.
- If every model on a category is unavailable simultaneously, check the help icon notifications for a platform-wide incident.
- A model in beta may move between tiers during rollout — the gate visible today may differ from the gate at GA.
What to do next
- Read the label on the unavailable model — it names the cause (provider, Martini-side, or plan).
- Switch to a sibling model from the node's picker if the work is time-critical.
- Check the help icon notifications for any active platform-level incident if multiple models are down.
- See the why-upgrade-prompt-shows article if the model is plan-gated on your current tier.
Related help articles
Related models and tools
Still need help? Contact support.