Anthropic Claude 4 Model Family

The current Claude 4 family comprises three generally available models plus a restricted research preview. Model selection is the first architectural decision in any Anthropic integration.

Current Model Lineup

Model	API ID	Alias	Context	Max Output	Pricing (input/output per MTok)
Claude Opus 4.7	`claude-opus-4-7`	`claude-opus-4-7`	1M tokens	128k (sync); 300k (batch)	$5 / $25
Claude Sonnet 4.6	`claude-sonnet-4-6`	`claude-sonnet-4-6`	1M tokens	64k (sync); 300k (batch)	$3 / $15
Claude Haiku 4.5	`claude-haiku-4-5-20251001`	`claude-haiku-4-5`	200k tokens	64k	$1 / $5

Batch 300k output requires the output-300k-2026-03-24 beta header. It is batch-only — unavailable on the synchronous Messages API.

Model Characteristics

Claude Opus 4.7 is the highest-capability generally available model. It offers a step-change improvement in agentic coding over Opus 4.6, 1M token context, and is the only model where adaptive thinking is the *only* supported thinking mode (manual budget_tokens returns 400).

Claude Sonnet 4.6 is the speed/intelligence balance point — fast, 1M context, supports both adaptive and manual thinking modes (manual is deprecated). It is the recommended default for most applications.

Claude Haiku 4.5 is the fastest model with near-frontier intelligence. 200k context, no adaptive thinking support — only manual extended thinking with budget_tokens.

Claude Mythos Preview (claude-mythos-preview) is a separate research preview for defensive cybersecurity, offered under Project Glasswing. Access is invitation-only with no self-serve sign-up. Adaptive thinking is its default and only supported mode; thinking.type: "disabled" is not supported; display defaults to "omitted".

Thinking Support Matrix

Model	Adaptive thinking	Manual thinking (`budget_tokens`)
Opus 4.7	Required	Rejected (400 error)
Sonnet 4.6	Recommended	Deprecated, functional
Opus 4.6	Recommended	Deprecated, functional
Haiku 4.5	Not supported	Supported
Mythos Preview	Default, only mode	Not applicable

Knowledge Cutoffs

Model	Reliable knowledge cutoff	Training data cutoff
Opus 4.7	Jan 2026	Jan 2026
Sonnet 4.6	Aug 2025	Jan 2026
Haiku 4.5	Feb 2025	Jul 2025

Retirement Dates

Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) — the earlier generation without the .6/.7 suffix — retire 2026-06-15. Migrate to Sonnet 4.6 or Opus 4.7 before that date.

Platform Availability

All models are available via the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. Bedrock offers global endpoints (dynamic routing) and regional endpoints (guaranteed data residency). Vertex AI offers global, multi-region, and regional endpoints.

US-only inference (inference_geo: "us") on Opus 4.7, Opus 4.6, and newer models bills at 1.1× the standard token rate. Priority Tier capacity is drawn down at 1.1 tokens per input/output token for US-only inference requests.

Note: Claude Opus 4.7 on AWS is available through Claude in Amazon Bedrock (the Messages-API Bedrock endpoint), not the Bedrock Converse API.

Models API

Query model capabilities and limits programmatically:

GET /v1/models

The response includes a capabilities object per model with fields: batch, citations, code_execution, context_management, effort, image_input, pdf_input, structured_outputs, thinking. This enables runtime capability checks rather than hardcoded model assumptions.

for model in client.models.list():
    if model.capabilities.thinking.types.adaptive.supported:
        print(f"{model.id} supports adaptive thinking")