# Claude Code Models, effort, and Fast Mode

> Claude Code: /model, /effort, opusplan, fallback models, Fast Mode, 1M context, provider capabilities, and caching.

URL: https://docs.passion8.cc/en/docs/claude-code/model-fast-mode
Language: en
Publisher: Passion8

Model configuration affects quality, speed, cost, and caching. With Passion8, verify the actual model ID, capability metadata, context window, effort, and Fast Mode support.

For advisor-assisted decisions, Agent View, goals, Ultraplan, and Ultrareview, see [Advanced workflows](https://docs.passion8.cc/en/docs/claude-code/advanced-workflows-cloud).

## Model selection precedence

| Source | Behavior | Use |
| --- | --- | --- |
| /model <alias or id> | Switch immediately; current versions can save a user default | Interactive selection |
| claude --model <alias or id> | This launch only | Different models in separate terminals |
| ANTHROPIC_MODEL | Sessions launched with that environment | Shell profiles or scripts |
| settings.json model | Persistent default | Personal/team preferences |
| Managed settings | Administrator policy | Organization restrictions |

Resuming a session normally uses its saved model unless unavailable, excluded by policy, or explicitly overridden by the new launch's model selection.




Switching model changes cache identity. The next turn must process history for the new model; neither five-minute nor one-hour TTL transfers it across models.




## Common aliases

| Alias | Meaning | Notes |
| --- | --- | --- |
| default | Account/provider/organization default | Varies by plan and provider |
| opus | Default Opus candidate | Planning, architecture, difficult review |
| sonnet | Default Sonnet candidate | Most coding execution |
| haiku | Default Haiku candidate | Lightweight work; capabilities vary |
| fable | Default Fable candidate; checked catalog lists Fable 5.1 | Safety/bio content may trigger fallback |
| opusplan | Opus for planning, Sonnet for execution | Mode transitions can switch model |
| opus[1m] / sonnet[1m] | Request a 1M-context variant | Requires model/account/provider support |

Use the actual Passion8 ID if the gateway uses custom names. For model discovery in the picker, see [Gateway and protocol](https://docs.passion8.cc/en/docs/claude-code/gateway).

## opusplan

| Phase | Model | Use |
| --- | --- | --- |
| Plan mode | Opus | Requirements, architecture, risk |
| Execution | Sonnet | Editing, testing, ordinary implementation |

The tradeoff is split caches when the actual model changes.

1. Plan at the beginning with opusplan.
2. After approval, avoid repeatedly moving between planning and execution.
3. For an occasional second opinion, consider advisor instead of switching the whole session.

## Effort

Effort controls adaptive reasoning depth. Higher settings can cost more and take longer; supported levels and defaults depend on the model.

| Effort | Typical use |
| --- | --- |
| low | Short, deterministic, latency-sensitive work |
| medium | Moderate tasks with cost constraints |
| high | Balanced coding work where supported |
| xhigh | Complex architecture, migration, debugging |
| max | Deep session work; avoid an unconditional global default |
| ultracode | Claude Code mode combining xhigh and more active dynamic workflows |

Set at launch:

```bash
claude --effort high
```

Or in-session:

```text
/effort xhigh
/effort auto
```

Or in settings:

```json title="settings.json"
{
  "effortLevel": "high"
}
```




Effort changes can invalidate cached history. Choose it before a long task. An explicit CLAUDE_CODE_EFFORT_LEVEL environment value takes precedence over session effort controls.




## Fast Mode

Fast Mode trades higher cost for lower latency on supported Opus configurations.

```text
/fast
```

| Scenario | Recommendation |
| --- | --- |
| Speed required from the start | Enable before the long task |
| Existing long session | Initial activation may require reprocessing history |
| Not currently Opus | Activation may switch model |
| Rate limited | Official service can fall back to standard speed; cache behavior depends on version |

Fast Mode is a speed/billing setting; effort controls reasoning depth. They are not interchangeable.

## Advisor tool

The main model continues working while consulting another model on difficult decisions.

```text
/advisor opus
/advisor off
```

At startup:

```bash
claude --advisor opus
```

| Aspect | Cache effect |
| --- | --- |
| Toggle advisor | Generally preserves main prefix |
| Advisor request | Reads the transcript; advisor calls do not share their own cache as ordinary continued turns |
| Advisor answer | Appended to main history and can participate in later caching |




Advisor is a server-executed Anthropic tool. Availability through cloud providers or third-party gateways depends on actual support and passthrough.




## Fallback models

Configure a fallback chain for overload/unavailability:

```bash
claude --fallback-model sonnet,haiku
```

Or in settings:

```json title="settings.json"
{
  "fallbackModel": ["sonnet", "haiku"]
}
```

A fallback turn uses another model and therefore another cache. It improves availability rather than reducing cache cost.

## 1M context

The official configuration supports [1m] aliases/suffixes on compatible models. Availability depends on model, account, provider, and gateway.

| Check | Reason |
| --- | --- |
| Actual 1M model route | An alias alone cannot create upstream support |
| Context/beta field passthrough | Stripping fields can disable the requested capability |
| Compaction threshold | A smaller gateway window may require CLAUDE_CODE_AUTO_COMPACT_WINDOW |

## Third-party capability declarations

Claude Code infers effort, thinking, tool search, and context capabilities from model IDs. Custom deployment names may need provider-specific declarations.

```bash
export ANTHROPIC_DEFAULT_OPUS_MODEL="my-opus-deployment"
export ANTHROPIC_DEFAULT_OPUS_MODEL_SUPPORTED_CAPABILITIES="effort,xhigh_effort,max_effort,thinking,adaptive_thinking,interleaved_thinking"
```

The official capability variables apply to supported provider configurations such as Bedrock, Vertex, Foundry, and Mantle. Do not assume the same behavior on every ANTHROPIC_BASE_URL gateway; full header/body forwarding remains essential.

## Passion8 recommendations

| Need | Recommendation |
| --- | --- |
| Stable coding | Available Sonnet/Opus ID; avoid frequent switching |
| Deep planning | Try opusplan or supported advisor separately |
| Long context | Verify model window and gateway fields before [1m] |
| Cost control | Choose a supported moderate effort; compact at natural breaks |
| Speed | Decide on Fast Mode before a long conversation |
| Gateway model picker | Requires /v1/models and CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1 |

## Official references

- [Model configuration](https://code.claude.com/en/docs/en/model-config.md)
- [Fast mode](https://code.claude.com/en/docs/en/fast-mode.md)
- [Advisor tool](https://code.claude.com/en/docs/en/advisor.md)
- [Prompt caching](https://code.claude.com/en/docs/en/prompt-caching.md)
- [Gateway protocol reference](https://code.claude.com/en/docs/en/llm-gateway-protocol.md)