Cmd. /model

Saves the selected model as the default for new sessions. Press s on a picker row, or use a supported -p invocation, for a session-only change.

Key takeaways

Select a model and save the default.
Check the input surface and availability before using the examples below.

Purpose

Select a model and save the default.

/model [model]

Availability

Field	Value
Input surface	Running CLI prompt
Kind	Built-in
Status	Current reference
Aliases	—
Documented minimum	Not specified
Reviewed against	2026-10-23 · Claude Code 2.1.308

Without arguments, opens a picker. In -p on v2.1.205+, supply a model; the change is session-only.

Examples

/model

Effects and persistence

Saves the selected model as the default for new sessions. Press s on a picker row, or use a supported -p invocation, for a session-only change.

Open /model to use the picker. A normal selection is saved as a default for new sessions; press s on a row for the current session only. Provider aliases can resolve to different model versions, so use a full model ID for reproducible comparisons.

Opus 5.5 Baseline

Claude Opus 5.5 (claude-opus-5-5) was released on 2026-09-22 and is the default Opus model from Claude Code v2.1.280 (also the minimum version). Official docs describe it as built for long-running agentic coding and knowledge work, and recommend starting most workloads on it.

Provider	`opus`	`sonnet`	`haiku`	`default`
Anthropic API (Pro, Max, Team, Enterprise, API keys)	Opus 5.5	Sonnet 5.5	Haiku 5.5	Opus 5.5
Claude Platform on AWS	Opus 5.5	Sonnet 4.6	Haiku 4.5	Opus 5.5
Amazon Bedrock, Google Cloud Agent Platform	Opus 5.5	Sonnet 4.5	Haiku 4.5	Opus 5.5
Microsoft Foundry	Opus 4.6	Sonnet 4.5	Haiku 4.5	Sonnet 4.5

Before v2.1.280, default was Sonnet 5 on Pro and Team Standard and Opus 5 elsewhere. Pro and Team Standard now default to Opus, matching Max, Team Premium, and Enterprise. An organization default model, when set, still replaces the account-type default.
Specs: 1M context, 128K max output, $4 input / $20 output per MTok, cache reads $0.20/MTok (5% of input), knowledge cutoff June 2026, retirement not sooner than 2027-09-22.
Default effort is medium, and a top-level effortLevel in user settings does not apply to Opus 5.5; pick a level with /effort or the /model picker. Thinking is always on, so the thinking toggle, alwaysThinkingEnabled, and MAX_THINKING_TOKENS=0 have no effect.
Fast mode defaults to Opus 5.5 from v2.1.280 at $8/$40 per MTok (Opus 5 and Opus 4.8: $10/$50).
Classifier fallback: Opus 5.5 biology flags rerun on Opus 5, cybersecurity flags on Opus 4.8. From v2.1.287 an automatic switch keeps the current effort level instead of the new model's default.
From v2.1.283 the picker's Opus row and the Default name drop "(1M context)"; the window is unchanged.
To hold new releases back, v2.1.283 adds managed deniedModels (block named models even when availableModels allows them) and availableModelsMatch: "exact" (each listed ID allows only that version). Older versions ignore both keys, so pair them with requiredMinimumVersion.
Messages API harnesses moving from Opus 5 must handle four breaking changes: thinking cannot be disabled, forced tool_choice (any/tool) returns 400, thinking blocks are tied to the model and conversation, and computer_20251124 is rejected on the Claude API and Google Cloud.
Opus 5 (claude-opus-5, v2.1.219+) is the previous default Opus. Pin it by full name for regression comparisons.

Sonnet 5.5 Baseline

Claude Sonnet 5.5 (claude-sonnet-5-5) was released on 2026-09-28 and is the default Sonnet model on the Anthropic API from Claude Code v2.1.284 (also the minimum version). On Claude Platform on AWS, Bedrock, Google Cloud Agent Platform, and Foundry the sonnet alias still resolves to the older models in the table above, so select Sonnet 5.5 by full name or set ANTHROPIC_DEFAULT_SONNET_MODEL there.

Specs: 1M context, 128K max output, $2 input / $10 output per MTok, cache reads $0.10/MTok (5% of input), knowledge cutoff June 2026, retirement not sooner than 2027-09-28.
On the Anthropic API it always runs with the 1M window: there is no 200K variant, sonnet[1m] has no effect, and no plan needs usage credits for it. The same window applies behind an LLM gateway or a custom ANTHROPIC_BASE_URL (v2.1.285); run /autocompact 200k if the gateway stops at 200K.
Default effort in Claude Code is medium, as on Opus 5.5, and thinking cannot be turned off.
Classifier fallback: cybersecurity flags rerun on Sonnet 5, and biology flags end with a refusal. On third-party providers the rerun needs ANTHROPIC_DEFAULT_SONNET_MODEL or a Sonnet 5 entry in the provider model list, in addition to the Opus pin.
From v2.1.288 the client-side auto mode classifier ignores an ANTHROPIC_DEFAULT_SONNET_MODEL pin that names Sonnet 5.5 or Opus 5.5 and uses Sonnet 5.
Sonnet 5 was the default Sonnet on the Anthropic API from v2.1.197 through v2.1.283. Pin it by full name for regression comparisons; its default effort is high.

Haiku 5.5 Baseline

Claude Haiku 5.5 (claude-haiku-5-5) was released on 2026-10-07 and is the default Haiku model on the Anthropic API from Claude Code v2.1.293; the Model configuration docs say to use v2.1.293 or later with it. On Claude Platform on AWS, Bedrock, Google Cloud Agent Platform, and Foundry the haiku alias still resolves to Haiku 4.5.

Select it with /model claude-haiku-5-5 or claude --model claude-haiku-5-5. When the model setting is haiku, a session saved on Haiku 4.5 resumes on Haiku 5.5, the model haiku resolves to now.
Specs: 1M context, 128K max output, knowledge cutoff June 2026, retirement not sooner than 2027-10-07. On the Anthropic API it runs with the 1M window on every plan with no [1m] suffix, and sessions auto-compact at about 967K tokens by default.
Pricing depends on prompt length. A request whose prompt is up to 100,000 tokens costs $0.10 input / $0.50 output per MTok; above that it costs $0.50 / $2.50. Cache reads are $0.01 ($0.05 above the threshold), 5-minute cache writes $0.125 ($0.625), 1-hour cache writes $0.20 ($1), and the Batch API is 50% off. Prompt length counts every input token, including cache reads and writes, and each request is priced on its own. To limit cost in long sessions, set a smaller window for this model with /autocompact so it compacts earlier.
Adaptive thinking is always on, so thinking cannot be turned off, as on Opus 5.5 and Sonnet 5.5. Default effort is medium, and the supported levels are low, medium, high, xhigh, and max.
It uses the same tokenizer as Claude 4.7 and later models, so the same text counts as about 30% more tokens than on Haiku 4.5.
v2.1.296 fixed Haiku 5.5 token counts that failed behind some gateways.

Fable 5.1 Baseline

Claude Fable 5.1 (claude-fable-5-1) was released on 2026-09-01 and is the default Fable model from Claude Code v2.1.257 (the model-config docs list v2.1.255 as the minimum). It is not the default model on any plan or provider; select it with /model fable, claude --model fable, the model setting, or best where the organization has access. Fable 5 remains available as a legacy model via /model claude-fable-5, and the availableModels prefix claude-fable-5 permits both.

Item	Fable 5.1
Pricing	$10 input / $50 output per MTok, same as Fable 5; cache reads $0.25/MTok (a quarter of Fable 5), 5m cache write $12.50, 1h cache write $20, Batch API 50% off
Limits	1M context (default and maximum), 128K max output, always-on adaptive thinking, effort `low` to `max` with `high` default
Retention	Covered Model with 30-day data retention; not available under ZDR unless Anthropic authorizes it
Classifier fallback	Fable 5.1/5 (and Opus 5.5) biology flags rerun on Opus 5, cybersecurity flags on Opus 4.8 (Opus 5 itself: cyber to Opus 4.8, biology refused); `switchModelsOnFlag: false` disables; third-party providers need `ANTHROPIC_DEFAULT_OPUS_MODEL`
Breaking API changes vs Fable 5	`tool_choice` `any`/`tool` return 400; thinking blocks are readable only by Fable 5.1 (older models drop them unbilled); editing earlier turns invalidates thinking blocks, enforced for accounts created on or after 2026-08-31
Additive API features	Per-message effort (`mid-conversation-output-config-2026-07-01` beta), turn-scoped system messages (`clear_at: "next_user_message"`, beta), `thinking.display: "updates"` (beta), content provenance
Mythos 5.1	Same specs and pricing, different safeguard level; Project Glasswing approved customers only, not selectable via `/model`

Claude Code keeps conversation history append-only, so the history-editing rule does not affect interactive sessions. Custom harnesses can start with /claude-api migrate this project to claude-fable-5-1.

Fable Plan Access

The temporary subscription allowance ended on 2026-07-07. As of 2026-07-19, keep seat-based subscription usage separate from API consumption. The Fable 5.1 announcement did not change plan inclusion, so treat these Fable 5 rules as the Fable-family baseline until a plan-specific notice lands:

Access path	Current rule	Operating note
Pro, Max, and Team	Usage credits required	Fable 5 is unavailable when credits are disabled or the balance is exhausted.
Standard Enterprise seat	Usage credits required	Standard seats were not part of the temporary included allowance.
Premium/select Enterprise seat	Usage credits required	Up to 50% of seat usage was included only through 2026-07-07.
Claude API and consumption-based Enterprise	Token-based API billing	Separate from weekly seat allowances; standard Fable 5 API pricing applies.

Usage credits are billed separately from the subscription at standard API rates. Configure payment, prepaid balance, monthly caps, and auto-reload in Settings > Usage; use /usage-credits from Claude Code to open the corresponding account flow. Teams should set an admin policy and spending cap before making Fable the default path.

Fable Refusal and Fallback

A Fable safety-classifier refusal is a successful HTTP 200 Messages API response on both Fable 5.1 and Fable 5. Branch on stop_reason, not on HTTP errors, empty content, or stop_details:

{
  "model": "claude-fable-5-1",
  "content": [],
  "stop_reason": "refusal",
  "stop_details": {
    "type": "refusal",
    "category": "cyber",
    "explanation": "..."
  }
}

stop_details.category may identify an area such as cyber or bio, but it can be null and the human-readable explanation is not a stable parsing contract.

Fallback path	Availability	Implementation rule
Server-side	Beta on the Claude API and Claude Platform on AWS	Pass up to three models in `fallbacks` with the `server-side-fallback-2026-06-01` beta header, or use `fallbacks: "default"` with `server-side-fallback-2026-07-01` to retry on Anthropic's recommended model per category. Fable 5.1 permits Opus 4.8 and Opus 5 as targets.
SDK middleware	Supported platforms	Reuse conversation fallback state so follow-up turns stay pinned to the model that accepted the request.
Manual retry	Any platform or language	Detect the refusal, retry on another model, and redeem the `fallback-credit-2026-06-01` credit token.

Do not combine server-side fallbacks and the SDK refusal-fallback middleware on one request.
Message Batches can return refused items as result.type: "succeeded"; inspect stop_reason. Server-side fallbacks is not supported for batches.
Claude Code fallbackModel and --fallback-model handle availability failures. They are not the same setting as the Messages API refusal fallback.
A request refused before output is not billed. Server-side fallback and the SDK middleware apply fallback credit automatically; manual retry logic must redeem the credit token to avoid paying the prompt-cache write cost twice when switching models.
Instrument refusals and fallback-served responses separately. HTTP-error monitoring alone cannot reveal this path.

Prompts, skills, and harness instructions that ask Fable to echo, transcribe, or expose its internal reasoning can trigger the reasoning_extraction refusal category and increase fallback to Opus 4.8. Ask for a concise rationale, verification evidence, and user-facing progress instead. When an API integration needs reasoning visibility, consume the structured adaptive-thinking blocks.

Official Fable References

Claude Fable 5.1 overview: https://platform.claude.com/docs/en/models/fable-5-1/overview
What's new in Claude Fable 5.1: https://platform.claude.com/docs/en/models/fable-5-1/whats-new-fable-5-1
Migrating to Claude Fable 5.1: https://platform.claude.com/docs/en/models/fable-5-1/migration-guide
Prompting Claude Fable 5.1: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1
Fable 5.1 and Mythos 5.1 announcement: https://www.anthropic.com/claude-fable-and-mythos-5-1
Fable 5 introduction and API differences: https://platform.claude.com/docs/en/models/fable-5/introducing-claude-fable-5-and-claude-mythos-5
Refusals and fallback: https://platform.claude.com/docs/en/build-with-claude/refusals-and-fallback
Fallback credit: https://platform.claude.com/docs/en/build-with-claude/fallback-credit
Prompting Claude Fable 5: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5
Usage credits: https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans

Cautions

Switching with existing conversation history may require a full uncached reread. Check the selected model and provider after switching. From v2.1.287, Opus 4.7 and later and Fable use a 1M context window by default on Bedrock, Vertex, Foundry, and the Claude apps gateway with no [1m] suffix; CLAUDE_CODE_DISABLE_1M_CONTEXT=1 keeps 200K.

Official sources

Complete command list · Verification scope

Key takeaways

Select a model and save the default.
Check the input surface and availability before using the examples below.

Purpose

Select a model and save the default.

/model [model]

Availability

Field	Value
Input surface	Running CLI prompt
Kind	Built-in
Status	Current reference
Aliases	—
Documented minimum	Not specified
Reviewed against	2026-10-23 · Claude Code 2.1.308

Without arguments, opens a picker. In -p on v2.1.205+, supply a model; the change is session-only.

Examples

/model

Effects and persistence

Saves the selected model as the default for new sessions. Press s on a picker row, or use a supported -p invocation, for a session-only change.

Choosing scope

Opus 5.5 Baseline

Provider	`opus`	`sonnet`	`haiku`	`default`
Anthropic API (Pro, Max, Team, Enterprise, API keys)	Opus 5.5	Sonnet 5.5	Haiku 5.5	Opus 5.5
Claude Platform on AWS	Opus 5.5	Sonnet 4.6	Haiku 4.5	Opus 5.5
Amazon Bedrock, Google Cloud Agent Platform	Opus 5.5	Sonnet 4.5	Haiku 4.5	Opus 5.5
Microsoft Foundry	Opus 4.6	Sonnet 4.5	Haiku 4.5	Sonnet 4.5

Before v2.1.280, default was Sonnet 5 on Pro and Team Standard and Opus 5 elsewhere. Pro and Team Standard now default to Opus, matching Max, Team Premium, and Enterprise. An organization default model, when set, still replaces the account-type default.
Specs: 1M context, 128K max output, $4 input / $20 output per MTok, cache reads $0.20/MTok (5% of input), knowledge cutoff June 2026, retirement not sooner than 2027-09-22.
Default effort is medium, and a top-level effortLevel in user settings does not apply to Opus 5.5; pick a level with /effort or the /model picker. Thinking is always on, so the thinking toggle, alwaysThinkingEnabled, and MAX_THINKING_TOKENS=0 have no effect.
Fast mode defaults to Opus 5.5 from v2.1.280 at $8/$40 per MTok (Opus 5 and Opus 4.8: $10/$50).
Classifier fallback: Opus 5.5 biology flags rerun on Opus 5, cybersecurity flags on Opus 4.8. From v2.1.287 an automatic switch keeps the current effort level instead of the new model's default.
From v2.1.283 the picker's Opus row and the Default name drop "(1M context)"; the window is unchanged.
To hold new releases back, v2.1.283 adds managed deniedModels (block named models even when availableModels allows them) and availableModelsMatch: "exact" (each listed ID allows only that version). Older versions ignore both keys, so pair them with requiredMinimumVersion.
Messages API harnesses moving from Opus 5 must handle four breaking changes: thinking cannot be disabled, forced tool_choice (any/tool) returns 400, thinking blocks are tied to the model and conversation, and computer_20251124 is rejected on the Claude API and Google Cloud.
Opus 5 (claude-opus-5, v2.1.219+) is the previous default Opus. Pin it by full name for regression comparisons.

Sonnet 5.5 Baseline

Specs: 1M context, 128K max output, $2 input / $10 output per MTok, cache reads $0.10/MTok (5% of input), knowledge cutoff June 2026, retirement not sooner than 2027-09-28.
On the Anthropic API it always runs with the 1M window: there is no 200K variant, sonnet[1m] has no effect, and no plan needs usage credits for it. The same window applies behind an LLM gateway or a custom ANTHROPIC_BASE_URL (v2.1.285); run /autocompact 200k if the gateway stops at 200K.
Default effort in Claude Code is medium, as on Opus 5.5, and thinking cannot be turned off.
Classifier fallback: cybersecurity flags rerun on Sonnet 5, and biology flags end with a refusal. On third-party providers the rerun needs ANTHROPIC_DEFAULT_SONNET_MODEL or a Sonnet 5 entry in the provider model list, in addition to the Opus pin.
From v2.1.288 the client-side auto mode classifier ignores an ANTHROPIC_DEFAULT_SONNET_MODEL pin that names Sonnet 5.5 or Opus 5.5 and uses Sonnet 5.
Sonnet 5 was the default Sonnet on the Anthropic API from v2.1.197 through v2.1.283. Pin it by full name for regression comparisons; its default effort is high.

Haiku 5.5 Baseline

Select it with /model claude-haiku-5-5 or claude --model claude-haiku-5-5. When the model setting is haiku, a session saved on Haiku 4.5 resumes on Haiku 5.5, the model haiku resolves to now.
Specs: 1M context, 128K max output, knowledge cutoff June 2026, retirement not sooner than 2027-10-07. On the Anthropic API it runs with the 1M window on every plan with no [1m] suffix, and sessions auto-compact at about 967K tokens by default.
Pricing depends on prompt length. A request whose prompt is up to 100,000 tokens costs $0.10 input / $0.50 output per MTok; above that it costs $0.50 / $2.50. Cache reads are $0.01 ($0.05 above the threshold), 5-minute cache writes $0.125 ($0.625), 1-hour cache writes $0.20 ($1), and the Batch API is 50% off. Prompt length counts every input token, including cache reads and writes, and each request is priced on its own. To limit cost in long sessions, set a smaller window for this model with /autocompact so it compacts earlier.
Adaptive thinking is always on, so thinking cannot be turned off, as on Opus 5.5 and Sonnet 5.5. Default effort is medium, and the supported levels are low, medium, high, xhigh, and max.
It uses the same tokenizer as Claude 4.7 and later models, so the same text counts as about 30% more tokens than on Haiku 4.5.
v2.1.296 fixed Haiku 5.5 token counts that failed behind some gateways.

Fable 5.1 Baseline

Item	Fable 5.1
Pricing	$10 input / $50 output per MTok, same as Fable 5; cache reads $0.25/MTok (a quarter of Fable 5), 5m cache write $12.50, 1h cache write $20, Batch API 50% off
Limits	1M context (default and maximum), 128K max output, always-on adaptive thinking, effort `low` to `max` with `high` default
Retention	Covered Model with 30-day data retention; not available under ZDR unless Anthropic authorizes it
Classifier fallback	Fable 5.1/5 (and Opus 5.5) biology flags rerun on Opus 5, cybersecurity flags on Opus 4.8 (Opus 5 itself: cyber to Opus 4.8, biology refused); `switchModelsOnFlag: false` disables; third-party providers need `ANTHROPIC_DEFAULT_OPUS_MODEL`
Breaking API changes vs Fable 5	`tool_choice` `any`/`tool` return 400; thinking blocks are readable only by Fable 5.1 (older models drop them unbilled); editing earlier turns invalidates thinking blocks, enforced for accounts created on or after 2026-08-31
Additive API features	Per-message effort (`mid-conversation-output-config-2026-07-01` beta), turn-scoped system messages (`clear_at: "next_user_message"`, beta), `thinking.display: "updates"` (beta), content provenance
Mythos 5.1	Same specs and pricing, different safeguard level; Project Glasswing approved customers only, not selectable via `/model`

Fable Plan Access

Access path	Current rule	Operating note
Pro, Max, and Team	Usage credits required	Fable 5 is unavailable when credits are disabled or the balance is exhausted.
Standard Enterprise seat	Usage credits required	Standard seats were not part of the temporary included allowance.
Premium/select Enterprise seat	Usage credits required	Up to 50% of seat usage was included only through 2026-07-07.
Claude API and consumption-based Enterprise	Token-based API billing	Separate from weekly seat allowances; standard Fable 5 API pricing applies.

Fable Refusal and Fallback

A Fable safety-classifier refusal is a successful HTTP 200 Messages API response on both Fable 5.1 and Fable 5. Branch on stop_reason, not on HTTP errors, empty content, or stop_details:

{
  "model": "claude-fable-5-1",
  "content": [],
  "stop_reason": "refusal",
  "stop_details": {
    "type": "refusal",
    "category": "cyber",
    "explanation": "..."
  }
}

stop_details.category may identify an area such as cyber or bio, but it can be null and the human-readable explanation is not a stable parsing contract.

Fallback path	Availability	Implementation rule
Server-side	Beta on the Claude API and Claude Platform on AWS	Pass up to three models in `fallbacks` with the `server-side-fallback-2026-06-01` beta header, or use `fallbacks: "default"` with `server-side-fallback-2026-07-01` to retry on Anthropic's recommended model per category. Fable 5.1 permits Opus 4.8 and Opus 5 as targets.
SDK middleware	Supported platforms	Reuse conversation fallback state so follow-up turns stay pinned to the model that accepted the request.
Manual retry	Any platform or language	Detect the refusal, retry on another model, and redeem the `fallback-credit-2026-06-01` credit token.

Do not combine server-side fallbacks and the SDK refusal-fallback middleware on one request.
Message Batches can return refused items as result.type: "succeeded"; inspect stop_reason. Server-side fallbacks is not supported for batches.
Claude Code fallbackModel and --fallback-model handle availability failures. They are not the same setting as the Messages API refusal fallback.
A request refused before output is not billed. Server-side fallback and the SDK middleware apply fallback credit automatically; manual retry logic must redeem the credit token to avoid paying the prompt-cache write cost twice when switching models.
Instrument refusals and fallback-served responses separately. HTTP-error monitoring alone cannot reveal this path.

Official Fable References

Claude Fable 5.1 overview: https://platform.claude.com/docs/en/models/fable-5-1/overview
What's new in Claude Fable 5.1: https://platform.claude.com/docs/en/models/fable-5-1/whats-new-fable-5-1
Migrating to Claude Fable 5.1: https://platform.claude.com/docs/en/models/fable-5-1/migration-guide
Prompting Claude Fable 5.1: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1
Fable 5.1 and Mythos 5.1 announcement: https://www.anthropic.com/claude-fable-and-mythos-5-1
Fable 5 introduction and API differences: https://platform.claude.com/docs/en/models/fable-5/introducing-claude-fable-5-and-claude-mythos-5
Refusals and fallback: https://platform.claude.com/docs/en/build-with-claude/refusals-and-fallback
Fallback credit: https://platform.claude.com/docs/en/build-with-claude/fallback-credit
Prompting Claude Fable 5: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5
Usage credits: https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans

Cautions

Official sources

Complete command list · Verification scope

Purpose

Availability

Examples

Effects and persistence

Choosing scope

Opus 5.5 Baseline

Sonnet 5.5 Baseline

Haiku 5.5 Baseline

Fable 5.1 Baseline

Fable Plan Access

Fable Refusal and Fallback

Official Fable References

Cautions

Official sources

On this page

Cmd. /model

Purpose

Availability

Examples

Effects and persistence

Choosing scope

Opus 5.5 Baseline

Sonnet 5.5 Baseline

Haiku 5.5 Baseline

Fable 5.1 Baseline

Fable Plan Access

Fable Refusal and Fallback

Official Fable References

Cautions

Official sources

On this page