Anthropic's Claude family includes Fable, Opus, Sonnet, and Haiku models. The best model ID for your application depends on the task, the provider you use, and the models your account can access. The quickest way to check your own Claude API access is the Models API: GET /v1/models. Anthropic's models overview is the reference for its current lineup.
This guide was checked against Anthropic documentation on November 10, 2026. Model availability, pricing, and lifecycle dates can change. Verify the latest information before deploying or migrating.
Current Claude model IDs
The following IDs are listed in Anthropic's current model comparison for the Claude API:
- Claude Fable 5.2 —
claude-fable-5-2: aimed at demanding reasoning and long-horizon agentic work. - Claude Opus 5.5 —
claude-opus-5-5: aimed at long-running agentic coding and knowledge work. - Claude Sonnet 5.5 —
claude-sonnet-5-5: positioned as a balance of speed and capability. - Claude Haiku 5.5 —
claude-haiku-5-5: intended for high-volume, latency-sensitive tasks such as classification and routing.
Anthropic describes these models as supporting text and image input, text output, multilingual use, vision, and tool use. The overview also lists legacy models that remain available, including earlier Fable, Opus, Sonnet, and Haiku releases. Availability and access can differ by organization and platform. See the official models overview for the current comparison and platform IDs.
How to list models available to your account
Use Anthropic's List Models endpoint. It returns model information in pages, with newer releases listed first. By default, the API includes active and deprecated models; retired models appear only when requested explicitly. Use limit to choose a page size and after_id or before_id for pagination.
A basic request is:
curl https://api.anthropic.com/v1/models \
-H 'anthropic-version: 2026-10-21' \
-H "Authorization: Bearer $ANTHROPIC_API_KEY"
Store your API key in a secret manager or environment variable; do not put it into source code. Consult the API reference for current authentication and version-header requirements before using the example. The endpoint response can include lifecycle, capability, context-window, and output-token information for each model. Check has_more and last_id when collecting a complete paginated list.
How to choose a Claude model
Start with the workload rather than a name alone:
- Define the task. A difficult, multi-step agent workflow may justify a reasoning-focused model; a high-volume classifier may benefit from a lower-latency model.
- Test representative inputs. Compare accuracy, reliability, latency, and total token usage on your own examples.
- Check required capabilities. The Models API exposes capability details, but organization settings can still affect whether a feature is usable.
- Verify costs and limits. Check current input and output pricing, context limits, rate limits, and your provider's terms. Do not assume a model listed in documentation is enabled for your account.
- Plan for change. Record the exact model ID in configuration, monitor lifecycle notices, and evaluate a replacement before retirement.
Anthropic's overview recommends Opus 5.5 as a starting point for many workloads and Fable 5.2 for more demanding reasoning and long-horizon agentic work. Those are starting points, not guarantees that either model is best for every application.
Claude API IDs versus cloud provider IDs
Model identifiers are platform-specific. For example, the Claude API identifies Opus 5.5 as claude-opus-5-5; Anthropic's overview lists its Amazon Bedrock ID as anthropic.claude-opus-5-5. A Microsoft Foundry deployment may use a deployment name configured by the customer. Confirm the exact identifier and endpoint with your chosen platform before sending requests. Availability and lifecycle schedules may also differ across providers.
How versioned model IDs work
According to Anthropic's model ID and versioning guide, Claude 4.6 and later use dateless IDs such as claude-sonnet-4-6. These IDs identify pinned snapshots; they are not automatic pointers to the newest release. Earlier models can use dated snapshot IDs and, in some cases, convenience aliases.
A pinned model ID does not freeze all serving infrastructure. Anthropic notes that routing, safety classifiers, and sampling infrastructure can change, so observable behavior may still vary. Check the official lifecycle documentation and test upgrades before changing production traffic.
Frequently asked questions
Is there a single complete Anthropic model list?
The models overview is useful for comparing the current lineup. To discover model information programmatically and check what your API account can use, query GET /v1/models and paginate the results. Include lifecycle filters if you need retired entries.
Which Claude model is cheapest?
Prices depend on the model, input and output token usage, service tier, and platform. Haiku is positioned for high-volume workloads, but check the current pricing page and your provider's terms instead of assuming it is cheapest for every workload.
Can I use the same model ID on every platform?
Not always. Check the provider-specific ID or deployment name in the official overview and your cloud platform's documentation.
Are all listed models available to my account?
No. Access can vary by account, organization settings, lifecycle stage, region, and provider. Confirm availability using the relevant provider's API and documentation.