Claude Sonnet 5 Economy
Sonnet 5 is Anthropic’s most capable Sonnet-class model, delivering frontier-level performance across coding, agentic workflows, and professional tasks. It features adaptive thinking with selectable reasoning levels—**low**, **medium**, **high**, **max**, and **x-high**—along with a 1-million-token context window and support for text, image, and file inputs. Built with an updated tokenizer, Sonnet 5 also incorporates real-time cybersecurity safeguards designed to block high-risk dual-use activities.
Related Models
Claude Fable 5.1 Economy
Claude Fable 5.1 is Anthropic's upgraded flagship frontier model engineered for high-complexity software development, long-horizon agentic workflows, and multi-step knowledge work. Built on Mythos-tier reasoning, Fable 5.1 significantly enhances autonomous self-verification and root-cause troubleshooting while reducing safety false-positive interventions by up to 60%. Powered by an optimized prompt-caching architecture featuring a 75% reduction in cache-read pricing, it cuts total operating costs by 25% to 45% for highly agentic tasks.
Claude Opus 5 Economy
Claude Opus 5 is Anthropic's ultra-flagship frontier model engineered for extreme cognitive density, deep logical reasoning, and expert-level scientific synthesis. Powered by native adaptive thinking and a 1-million-token context window, it resolves complex multi-step mathematical, architectural, and coding challenges with unprecedented accuracy and near-zero hallucination rates. Coupled with rigorous enterprise safety alignment, Claude Opus 5 excels at high-stakes operations, including legal compliance, quantitative financial modeling, advanced biosecurity research, and mission-critical software architecture auditing.
Claude Opus 4.8 Economy
Claude Opus 4.8 is Anthropic’s most powerful officially released Opus model, designed for complex reasoning, long-horizon agentic coding, and highly autonomous professional workflows. It supports text, image, and file inputs and produces text output. It features a 1-million-token context window, up to 128,000 output tokens, and built-in capabilities for adaptive thinking, tool use, and structured outputs. The model excels at advanced coding, browser and computer-use agents, enterprise knowledge work, financial and legal analysis, and multi-step tasks requiring sustained judgment and high reliability.
Claude Fable 5 Economy
Claude Fable 5 is Anthropic’s officially released Mythos-tier model, designed for autonomous knowledge work, advanced reasoning, and long-horizon coding. It supports text, image, and file inputs and produces text output. It features a 1-million-token context window, up to 128,000 output tokens, and built-in always-on adaptive thinking, tool use, and structured outputs. The model shares the same underlying model family as Claude Mythos 5 but was released publicly with stronger safety safeguards. It is suitable for complex software engineering, multi-step agentic workflows, enterprise research, document analysis, and professional tasks requiring sustained reasoning and high reliability.
Read Me
Claude Sonnet 5 API
Claude Sonnet 5 is Anthropic's most agentic Sonnet model at launch, designed to plan, use tools, execute multi-step work, and verify results with less scaffolding than earlier Sonnet generations. It combines a 1M-token context window, 128K maximum output, and default Adaptive Thinking for coding, professional knowledge work, and production agents.
On iCreat, you can call claude-sonnet-5 through our Anthropic-compatible Messages API. The endpoint supports text and image input, streaming, Adaptive Thinking, and thinking/text content blocks, giving existing Claude applications a familiar path to evaluate and deploy the model.
Model Positioning
Claude Sonnet 5 occupies Anthropic's balanced tier between speed-first Haiku models and capability-first Opus or Fable models. It is intended for teams that need stronger autonomous execution than Claude Sonnet 4.6 without routing every difficult request to a larger model.
Its main advantage is not a single benchmark result. It is the combination of agentic coding, tool-oriented reasoning, long-context capacity, and adjustable effort in a model intended to run at production scale. Use an Opus- or Fable-class model when the highest available capability matters more than latency; use Haiku when minimum response time is the dominant constraint.
Key Capabilities
Agentic Coding and Verification
Claude Sonnet 5 is built to carry software tasks through investigation, implementation, testing, and review. It is more likely than Sonnet 4.6 to use available tools, check its work, and continue until a multi-step change reaches a verifiable result.
Default Adaptive Thinking
Adaptive Thinking is enabled by default. The model decides whether and how much to reason for each request, while low, medium, high, xhigh, and max effort levels guide the balance between token use and capability. The API default is high.
Million-Token Context
The 1M-token context window supports large repositories, extensive document sets, long conversation histories, and multi-source analysis. A 128K maximum output provides room for substantial code, reports, plans, or structured deliverables.
Tool-Oriented Agent Workflows
Adaptive Thinking automatically enables Interleaved Thinking, allowing the model to reason between tool calls. With an application-provided tool layer, Sonnet 5 can work through browser, terminal, retrieval, or business-system tasks while preserving intermediate context.
Visual and Structured Work
Claude Sonnet 5 accepts text and image input and returns text. It can reason over screenshots, diagrams, charts, and document visuals, while structured outputs support workflows that require responses to follow a defined schema.
Pricing
| Token Type | Price |
|---|---|
| Input | $0.40 per 1M tokens |
| Output | $2 per 1M tokens |
| Cache Read | $0.04 per 1M tokens |
| Cache Write | $0.50 per 1M tokens |
Use Cases
- Production coding agents: investigate issues, edit code, run tests, and verify multi-file changes through application-provided tools.
- Codebase modernization: analyze large repositories and coordinate migrations, refactors, or dependency upgrades across related components.
- Professional knowledge work: synthesize large evidence sets into legal, financial, operational, or research deliverables for human review.
- Computer-use workflows: reason over screenshots and select actions in browser or desktop workflows when the application supplies a controlled computer-use environment.
- Long-context automation: process extensive documentation, support histories, policies, or records while maintaining instructions across multiple stages.
Consequential actions still require scoped permissions, application-level validation, audit logs, and human approval where appropriate.
Model Comparison
Claude Sonnet 5 vs Claude Sonnet 4.6
| Factor | Claude Sonnet 5 | Claude Sonnet 4.6 |
|---|---|---|
| Context Window | 1M tokens | 1M tokens |
| Maximum Output | 128K tokens | 128K tokens |
| Adaptive Thinking | On by default | Off unless enabled |
| Manual Extended Thinking | Not supported | Supported but deprecated |
| Sampling Parameters | Non-default values rejected | Non-default values accepted |
| Core Difference | Stronger coding and agentic execution | Previous balanced Sonnet generation |
Claude Sonnet 5 vs GPT-5.5 and Gemini 3.5 Flash
| Factor | Claude Sonnet 5 | GPT-5.5 | Gemini 3.5 Flash |
|---|---|---|---|
| Positioning | Balanced agentic coding and professional work | Frontier coding and complex professional work | Fast, scaled agentic and coding workflows |
| Context Window | 1M tokens | 1,050,000 tokens | 1,048,576 tokens |
| Maximum Output | 128K tokens | 128K tokens | 65,536 tokens |
| Official Input Modalities | Text and image | Text and image | Text, image, video, audio, and PDF |
| Reasoning Control | low, medium, high, xhigh, max; Adaptive Thinking |
none, low, medium, high, xhigh reasoning effort |
minimal, low, medium, high thinking level |
| Best Fit | Claude-based coding agents and long-context professional workflows | High-capability OpenAI agent and coding workflows | Broad multimodal inputs and rapid agentic loops |
Why Choose Claude Sonnet 5?
Choose Claude Sonnet 5 when a production application needs stronger planning, coding, tool use, and self-verification without defaulting to Anthropic's largest model. Its 1M context window and 128K maximum output make it suitable for tasks that span large source sets and produce substantial deliverables.
Default Adaptive Thinking also supports mixed request traffic. Straightforward work can stay comparatively direct, while difficult coding or agent tasks can use higher effort. This gives teams one Sonnet-tier model that can cover interactive use and deeper multi-step execution with request-level control.
Specifications
| Category | Description |
|---|---|
| Model Name | Claude Sonnet 5 |
| Provider | Anthropic |
| Model ID | claude-sonnet-5 |
| Release Date | June 30, 2026 |
| Model Type | Hybrid-reasoning LLM |
| Context Window | 1M tokens |
| Maximum Output | 128K tokens |
| Reliable Knowledge Cutoff | January 2026 |
| Training Data Cutoff | January 2026 |
| Official Input Modalities | Text and image |
| Output Modalities | Text |
| Official Reasoning Control | Adaptive Thinking; low, medium, high, xhigh, max effort |
| Default Effort | high |
| iCreat API Features | Anthropic-compatible Messages API, text and image input, streaming, Adaptive Thinking, thinking/text content blocks |
| Best Suited For | Agentic coding, tool workflows, long-context analysis, professional knowledge work |
Architecture
Anthropic does not publish Claude Sonnet 5's parameter count or underlying network architecture. Its documented behavior is that of a hybrid-reasoning model: Adaptive Thinking can answer directly or allocate more reasoning based on request complexity and the selected effort level.
Sonnet 5 also introduces a new tokenizer. The same text produces approximately 30% more tokens than on Sonnet 4.6, although the exact change depends on the content. This changes practical context capacity, output sizing, and usage estimates even when the surrounding API request shape remains the same.
Production Notes
Recount representative prompts with Sonnet 5 before migration. Do not reuse token estimates, max_tokens values, or context packing thresholds measured with Sonnet 4.6, because the new tokenizer can change how much text fits within the same nominal token limit.
At higher effort levels, Adaptive Thinking may consume a substantial part of the output budget. Leave enough headroom in max_tokens for both reasoning and the final answer, and monitor stop_reason: "max_tokens" during evaluation.
Preserve returned thinking blocks unchanged when continuing tool-use conversations. Stable thinking modes help preserve message-level cache breakpoints, while switching between adaptive and disabled thinking can break those message cache points.
Evaluate complete task trajectories rather than isolated answers. Measure tool errors, retries, verification success, latency, token use, and human corrections before routing production work to the model.
FAQ
Why is the thinking field empty even though Adaptive Thinking is enabled?
Claude Sonnet 5 defaults thinking.display to omitted. The response can still contain a thinking block and signature while its readable text is empty. Request display: "summarized" when a reasoning summary is needed, and never treat that summary as raw chain-of-thought.
Can an existing Sonnet 4.6 request keep using budget_tokens?
No. Manual Extended Thinking was removed in Sonnet 5 and the request returns a 400 error. Use Adaptive Thinking with an effort level instead.
Can temperature, top_p, or top_k be tuned on Claude Sonnet 5?
Non-default values return a 400 error. Use the default sampling behavior and control tone, scope, and variation through clear system and user instructions.
How should an application detect a cybersecurity refusal?
A protected request may return HTTP 200 with stop_reason: "refusal" rather than an HTTP error. Check the response stop reason explicitly and provide an appropriate product-level fallback or user message.