Claude Sonnet 5.5: Pricing, Specs & 1M Context Window

Anthropic has introduced Claude Sonnet 5.5 in a breaking release deploying simultaneously across major cloud platforms and developer environments, arriving as the second model in the Claude 5.5 family following Opus 5.5. The release brings a 1-million-token native context window, default adaptive thinking, and immediate ecosystem availability. Claude Sonnet 5.5 is designed for focused coding, knowledge work, and well-scoped everyday work like building features and fixing bugs. According to Anthropic’s release materials, Claude Sonnet 5.5 is a clear upgrade over Sonnet 5, running more than 30% faster and costing up to 30% less for most workloads. Early testing by Epic Games shows that Claude Sonnet 5.5 cleared the expected quality bar on a system design audit and a data-flow review, managed tens of thousands of lines of code for gameplay system architecture, kept responses snappy, handled multi-hour tasks, and delivered with less prescriptive prompting. Daniel Vogel, COO at Epic Games, reported that Claude Sonnet 5.5 held up on complex architectural checks and maintained high throughput during extended development tasks. Furthermore, Hacker News community documentation and official release notes highlight that Sonnet 5.5 is positioned as a faster, lower-cost complement to Opus 5.5, which remains built for complex work requiring careful judgment and long-horizon reasoning. In contrast, Sonnet 5.5 excels at well-scoped everyday tasks, rapid feature iteration, bug fixing, and the generation of polished documents, slides, and spreadsheets backed by a strong eye for design. Additional evaluations show that Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, an agentic coding evaluation, compared to Sonnet 5 at 10.3%. On frontier coding and knowledge work evaluations, the model demonstrates significant improvements across various professional domains, while also achieving notable milestones in visual understanding as the first Sonnet model to complete complex interactive challenges working solely from raw screenshots.

Platform Availability and Ecosystem Integration

Cloud infrastructure providers and developer tooling ecosystems deployed Claude Sonnet 5.5 instantly. Amazon Bedrock, Claude Platform on AWS, and GitHub Copilot ingest the model for immediate everyday workflows. AWS AI highlights the model’s lower cost per task at faster speed. Broader availability spans the Claude API, Google Cloud, and Microsoft Foundry on Global Standard deployments, as well as wider access across Amazon Web Services, Google Cloud, and Microsoft Azure. Within the developer ecosystem, Claude Code incorporates support starting from version v2.1.284, where the standard sonnet alias resolves directly to Claude Sonnet 5.5 on the Claude API. Developers initiate work on the Claude Platform using the model identifier claude-sonnet-5-5. Meanwhile, Claude Haiku 5.5 joins the Claude 5.5 family in the coming weeks to target high-volume, low-latency workflows. Across these deployments, Claude Sonnet 5.5 maintains rate limits separate from Sonnet 5 while sharing identical default tier values.

Technical Specifications and Context Limits

Claude Sonnet 5.5 introduces advanced engineering capabilities for developers handling large-scale data and extensive codebases. The model features a native context window of 1 million tokens without requiring any special beta header. For output generation, Claude Sonnet 5.5 supports a maximum output of 128k tokens, which expands up to 300k tokens on the Message Batches API when using the output-300k-2026-03-24 beta header. The knowledge cutoff for Claude Sonnet 5.5 extends to June 2026. Regarding model identification, the model ID is claude-sonnet-5-5 on the Claude API, Claude Platform on AWS, Google Cloud, and Microsoft Foundry, whereas Amazon Bedrock designates the model as anthropic.claude-sonnet-5-5. For multimedia processing, Claude Sonnet 5.5 utilizes the high-resolution image tier, which supports images up to 2576 pixels on the long edge. These specifications enable engineering teams to ingest massive repositories and detailed visual assets into a single prompt context without encountering traditional token truncation bottlenecks.

Pricing Structure and Caching Costs

While the per-token price for Claude Sonnet 5.5 remains unchanged compared to Sonnet 5, the model typically requires fewer tokens to complete identical tasks, resulting in cost reductions of up to 30% for most operations. The standard input price for Claude Sonnet 5.5 is set at $2 per million tokens. The output price for Claude Sonnet 5.5 is $10 per million tokens. For prompt caching mechanisms, the cache read price is $0.20 per million tokens. Cache write pricing varies by retention duration, with the 5-minute cache write price established at $2.50 per million tokens and the 1-hour cache write price set at $4 per million tokens. The minimum cacheable prompt size for Claude Sonnet 5.5 is 512 tokens. Furthermore, US-only inference carries an additional multiplier, costing 1.1 times the standard price for Claude Sonnet 5.5. These transparent pricing tiers allow development teams to forecast expenditure accurately while scaling automated coding agents and continuous knowledge extraction pipelines.

Adaptive Thinking and Tool Use Configurations

Claude Sonnet 5.5 features adaptive thinking enabled by default, supporting configurable effort levels including low, medium, high, xhigh, and max. The default effort level is set to high on the Claude API and medium when operating within Claude Code. Because the model thinks by default, downstream application code must account for response structures that may begin with a thinking block before delivering textual output. For advanced function calling, strict tool use with Claude Sonnet 5.5 requires the additionalProperties parameter to be explicitly set to false on every object schema. In addition, platform integrations dictate specific tooling versions; on the Claude API and Google Cloud, Claude Sonnet 5.5 supports computer use exclusively through computer_toolset_20260801, whereas Amazon Bedrock continues to accept the legacy computer_20251124 identifier. When utilizing the advisor tool with a Sonnet 5.5 executor, older models such as Opus 4.8, Opus 4.7, and Sonnet 5 are rejected as advisors. Conversely, accepted advisors include Opus 5.5, Opus 5, and Sonnet 5.5 itself, with advice returned securely and encrypted inside an advisor_redacted_result block. Finally, Anthropic notes that Claude’s thinking process cannot be decoupled from the specific account that created it, preserving complete auditability across multi-agent workflows.

Security Safeguards, Refusals, and Compliance

Claude Sonnet 5.5 is the first Sonnet model to incorporate cybersecurity safeguards comparable to Anthropic’s most capable frontier models. In practical operation, a declined request for Claude Sonnet 5.5 returns an HTTP 200 status code with the stop_reason field set to refusal and stop_details naming one of five specific policy categories. To maintain workflow continuity, server-side fallback on the Claude API automatically retries cyber and frontier LLM declines on Sonnet 5. In addition to behavioral guardrails, Claude Sonnet 5.5, along with Opus 5.5 and Sonnet 5, is available with zero data retention for eligible enterprise customers. Expanding upon these security measures, Anthropic announced that the Cyber Verification Program will soon expand to include Claude Sonnet 5.5 for authorized security research and legitimate defensive work.