> ## Documentation Index
> Fetch the complete documentation index at: https://evolink.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Models, costs, and quotas
> Shared model selection, estimate states, approval, access, and budget boundaries
This page maintains business rules shared by CLI and MCP. See [CLI reference](/docs/en/cli/reference) and [MCP reference](/docs/en/mcp/tools) for exact inputs. It describes released CLI 0.8.1 and MCP 1.6.1.
## Choose a model
1. Identify the task: image generation/editing, text/image-to-video, music, or speech. Specify count, quality, duration, references, and budget.
2. Query the account-available catalog, use its exact model ID, and inspect the schema. Family names, aliases, and display names are not automatically submission IDs.
3. Compare suitable newer or recommended families, then explain quality, reference support, cost, and expected wait. Do not claim measured popularity without platform statistics.
4. Check required fields, ranges, and mutually exclusive inputs before estimating the final request. Reconfirm after changing models or inputs.
Families worth comparing include GPT Image 2.5/2.0 and Seedream 5.0 for images; Seedance 2.5/2.0 and Wan 3.0 for video; Suno v6 for music. These are search directions, not guarantees of account availability. Choose speech models for narration. Example models illustrate workflows rather than a default recommendation.
| Information | CLI | MCP |
| - | - | - |
| Available models | `models search`, `models recommend` | `search_models`, `recommend_models` |
| Parameters and limits | `models show`, `models schema` | `get_model` |
| Model documentation | `docs search` | `search_docs` |
| Estimated cost for final inputs | `estimate` | `estimate_cost` |
## Read an estimate
| Status | Meaning | Action |
| - | - | - |
| `estimated` | Estimated amount; actual settlement still applies | Present the plan, amount, and known limitations, then obtain approval |
| `partial` | Missing pricing factors | Do not treat as a complete total; identify and resolve missing factors |
| `token_billed` | Billed by actual tokens | Explain metering rather than a fixed total |
| `needs_input` | Required input missing | Supply duration, references, or other required information and estimate again |
| `no_price` | No usable price | Do not describe as free or submit automatically |
Check `input_valid`, `problems`, `basis`, and `warnings`. Catalog starting prices are not complete request quotes. Resolution, sound, references, actual input duration, and audio billing methods can affect costs.
Released clients mainly use public default prices, not complete account discounts, measured usage, or account-settlement Quotes. New endpoints or code candidates do not establish production adoption. Pause when pricing or budget comparison is incomplete. The reference conversion is about 68 credits per \$1; use actual responses and account records.
## Approval and budgets
Show the exact model, final inputs, output specifications/count, estimated cost, and limitations. Obtain explicit approval for this generation. Changes to model, references, inputs, identity, price, or an expired quote require a new estimate and approval.
| Constraint | Current meaning |
| - | - |
| Account sign-in | Access to account services |
| Host command/tool permission | Permission for the agent to execute an operation |
| Per-task cost approval | Acceptance of this plan and estimated cost |
| Explicit `max_cost_usd` | Pre-submission estimate comparison in released clients; **not a final charge cap** |
| Platform Key quota | Account daily/total limits and access policies |
These are separate checks. Sign-in and host permission do not approve a charge. Do not remove budgets, change Keys, or switch routes to bypass rejection. Final caps require corresponding server support; current documentation cannot promise settlement never exceeds an estimate budget.
CLI stores a local `quote_id` and submits with `--confirm`. Native MCP keeps inputs, estimates, and explicit approval in the agent workflow; it has no CLI quote file or session-level server approval button. See [CLI approval](/docs/en/cli/billing#pricing) and [MCP approval](/docs/en/mcp/billing#pricing).
## Permissions, balance, and quotas
Same-account CLI and remote MCP OAuth can currently share internal Key permissions, quotas, and paused status. Changes affect its consumers, while CLI and host sessions remain separate. Local stdio uses its configured personal API Key.
Model, balance, and task queries do not submit media generation. Generation is billed; uploads have separate file-service limits. Resolve insufficient funds, quota, or access in the relevant account settings. Reauthentication cannot add funds or unpause a Key.
Failure, timeout, sign-out, or closing a page do not establish no charge or a refund. Report billing conclusions only from explicit evidence; usage queries are not complete statements.
## Empty search, invalid inputs, or incomplete pricing
1. Broaden wording while retaining required task/reference capabilities; check account and permissions.
2. Read this model's schema again and check fields, enums, ranges, and mutually exclusive references.
3. Estimate final inputs and follow the status table. Do not fill unknown factors with starting prices.
4. Reconfirm requests not yet submitted. For submitted or unknown requests, [recover first](/docs/en/cli-mcp/tasks-and-files#recovery).
See [CLI model troubleshooting](/docs/en/cli/billing#models-and-pricing) and [MCP model troubleshooting](/docs/en/mcp/billing#models-and-pricing) for route-specific actions.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.