Models

Model identifiers use the form providerID/modelID.

Select A Model

• UI: model picker
• TUI: /models
• CLI flag: --model provider/model

Default Model Fields

• model: primary model
• small_model: lightweight helper model for low-cost background tasks

{
  "model": "<provider>/<model-id>",
  "small_model": "<provider>/<small-model-id>"
}

Resolution Order

For an interactive client, model selection preference is normally:

1. explicit CLI/UI selection (--model or picker)
2. config model
3. a model restored from the active session, when that client supports session restoration
4. provider fallback ordering

For headless/default resolution, the runtime uses model when configured and otherwise chooses an available model from the allowed provider inventory. There is no universal model that is guaranteed to exist, be free, or be usable without provider setup.

BYOK And Cost Tiers

AlphaBase is bring-your-own-key (BYOK): you connect provider credentials, and that provider bills you directly for usage. AlphaBase does not mark up or resell model access.

AlphaBase does not guarantee a free model or a bundled usage allowance. Some provider catalogs expose no-cost models, and local runtimes may avoid hosted API charges, but availability, performance, infrastructure cost, and provider terms are configuration-dependent. Use the active model selector and provider documentation as the source of truth. See Providers.

For complex multi-step or multi-file tasks, validate the model you connect on real work before relying on it as your default. Use Launch Gates (certify:model) to measure that behavior.

Model Reliability And Capability Learning

AlphaBase persists per-model attempt evidence across sessions in provider/model-capability-health. The public certification states are:

State	Meaning
untested	The model is listed or is being checked, but has no current usable live evidence.
ready	A recent live request completed successfully, or the local runtime reported the model ready.
degraded	Recent evidence is limited, mixed, stale, or rate-limited; AlphaBase should recheck it.
unavailable	Authentication, missing setup, or the runtime/provider made the model unavailable.

These are evidence states, not a permanent quality ranking. The evidence includes success/failure, empty completions, suspicious provider events, latency, tool calls, and capability downgrades. When health-aware Smart Router is enabled, that evidence can affect fallback selection; it does not turn a catalog entry into a guarantee of model quality.

Idle Timeouts For Reasoning Models

The runtime applies model-specific stream deadlines. MiniMax cloud-Ollama models have dedicated bounded timeout knobs; those knobs are documented in Config. Timeout tuning does not provide provider access or imply that a model is free.

Smart Router (Opt-In, Advanced)

AlphaBase includes an internal Smart Router that can rank candidate models for an attempt and fail over to another model on a stuck/failed attempt. It is disabled by default and only engages when explicitly enabled for that request — AlphaBase never silently switches your selected model mid-task.

Smart Router is currently configured per-request via the prompt’s agentOptions (an SDK/API-level option), under any of these equivalent keys: modelRouter, modelRouting, or smartRouting. It is not yet a static alphabase.json top-level key.

// agentOptions passed with a prompt request
{
  "modelRouter": {
    "enabled": true,
    "models": ["<provider>/<model-id>", "<fallback-provider>/<model-id>"],
    "strategy": "balanced",
    "maxAttempts": 3,
    "cooldownSeconds": 30,
    "sameProviderOnly": false,
    "pinModelPerTurn": false,
    "respectRankOrder": true,
    "healthAwareRouting": true,
  },
}

Notes:

• sameProviderOnly filters the candidate pool to the requested provider. respectRankOrder controls whether the listed order wins over strategy scoring. pinModelPerTurn is an execution-contract option for clients that need a stable model across a turn; do not assume it overrides every host’s retry policy.
• Attempt 1 honors the requested/ranked pool. Failover selection is considered only for later attempts, up to the configured maxAttempts value, and may be limited by compatibility, health, and cooldown checks.
• There are no supported environment switches that globally force Smart Router or retry failover. Enable it explicitly in the request options above.

Per-Model Overrides

provider.<id>.models.<modelID> supports options and variants.

{
  "provider": {
    "<provider-id>": {
      "models": {
        "<model-id>": {
          "options": {
            "timeout": 300000,
          },
          "variants": {
            "high": { "disabled": false },
            "max": { "disabled": false },
          },
        },
      },
    },
  },
}

Variant Notes

• Variants are provider/model-specific.
• You can disable unwanted variants with variants.<name>.disabled.
• Agent-level model + variant settings can override global defaults.

Provider/Model Discovery

Use:

• GET /provider
• GET /provider/capability-index
• GET /provider/model-search?q=...

for capability-aware model routing and diagnostics.