Anthropic/Model/Claude/4.7/Opus
Key Features
- Generally available as of April 2026; positioned as Anthropic's most capable generally available model for complex reasoning and agentic coding
- Context window: 1M input tokens at standard API pricing (no separate long-context surcharge on the Messages API)
- Max output: 128k tokens on the synchronous Messages API (higher limits on Message Batches API with beta headers per Anthropic docs)
- Adaptive thinking only:
thinking: {"type": "adaptive"} is the supported thinking-on mode; extended thinking with explicit budget_tokens is not supported and returns HTTP 400 - New
xhigh effort level between high and max on the Messages API for finer latency vs. depth tradeoffs - High-resolution vision: up to 2576px on the long edge (~3.75MP); image coordinates map 1:1 to pixels for tool and UI workflows
- Task budgets (beta): advisory per-loop token budgets via beta header
task-budgets-2026-03-13 and output_config.task_budget - New tokenizer vs. prior Opus models: expect roughly 1.0x-1.35x text token counts vs. Opus 4.6 for similar content; validate with
/v1/messages/count_tokens - Sampling parameters: non-default
temperature, top_p, and top_k return HTTP 400; omit them and steer behavior in prompts - Cyber safeguards: automated refusals on prohibited or high-risk cybersecurity requests; legitimate security workflows may require Anthropic's Cyber Verification Program
- Reliable knowledge cutoff: January 2026 (per Anthropic models overview); training data cutoff: January 2026
- Model version status: Stable (general availability)
Performance Highlights
- Strong gains on difficult software engineering and long-horizon agentic work vs. Claude Opus 4.6 in Anthropic's public messaging and partner evaluations
- Improved knowledge work where the model must visually verify its own outputs (documents, slides, charts)
- Better use of filesystem-style memory and scratchpads across multi-turn agent runs
Model Tiers
- Flagship Opus release in the Claude 4 generation
- Earlier Opus pages in this garden:
- Speed-focused sibling line (separate page in graph when curated): Claude Sonnet 4.6 (
claude-sonnet-4-6 in Anthropic's model table)
Access Details
- API availability: Claude Messages API model string
claude-opus-4-7 (alias resolves to the current snapshot; see Anthropic's models overview for snapshot-pinned IDs) - Pricing (standard tier, per million tokens): $5 input ยท $25 output (unchanged vs. Opus 4.6 per Anthropic announcement)
- Platforms: Claude consumer and enterprise plans, Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry
- Amazon Bedrock base model ID:
anthropic.claude-opus-4-7 (plus global., us., eu., and jp. inference routing variants per AWS model card) - Google Vertex AI ID:
claude-opus-4-7 (see Google Cloud model registry for regional endpoint strings) - SDK support:
- Python:
pip install -U "anthropic" - TypeScript:
@anthropic-ai/sdk (and provider-specific Bedrock helpers per Anthropic's Bedrock integration docs)
Technical Specifications
- Model architecture: Proprietary Anthropic large language model (undisclosed weights)
- Input/output formats: Text and images in; text (and tool use) out; same tool-use platform as Opus 4.6 aside from thinking and sampling changes above
- Special capabilities:
- Adaptive thinking with optional
display: "summarized" to stream or summarize reasoning blocks (thinking blocks may be empty by default) - Task budgets beta for agentic loops
- High-resolution image understanding and improved low-level perception tasks
- Model version code:
claude-opus-4-7 (API alias; consult Anthropic's models overview for dated snapshot IDs used for cross-provider reproducibility) - Safety measures: Automated cybersecurity refusal classifiers plus Anthropic's broader safety stack; stricter handling of high-risk cyber prompts than prior Opus releases
References