Claude Sonnet 5.5 Official Specifications, Pricing & Benchmarks
Most popular daily coding & agent model worldwide ($2/$10) with 1M context and 128K output.
📋 Claude Sonnet 5.5 Official API & System Card Specifications
Verified against official Anthropic developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- claude-sonnet-5-5
- Release Date
- 2026-09-28
- Weight License
- Proprietary (Undisclosed)
- Native Reasoning
- ✅ Supported
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 1,000,000 tokens (1000K)
- Max Output Tokens
- 128,000 tokens
- Input Modalities
- Text + Image + PDF
- Output Modalities
- Text
- Direct China Access
- 🌐 Global / OpenRouter
- Deployment Mode
- Cloud API only (Proprietary)
3. Official Pricing & Cache
USD / 1M- Input / 1M Tokens
- $ 2
- Cached Input / 1M
- $ 0.1 -95%
- Output / 1M Tokens
- $ 10
- Tiered Pricing Notes
- 1M context window; $0.10/1M cache read; default daily driver across AI IDEs
🏆 Official Benchmark Results
Sourced from official System Card / Model Card & technical reports
A verified subset of 500 software engineering problems from real GitHub issues, validated by human annotators for evaluating language models' ability to resolve real-world coding issues by generating patches for Python codebases.
PhD-level Google-proof science reasoning questions
Dual-control conversational tool-use benchmark
A more robust and challenging multi-task language understanding benchmark that extends MMLU by expanding multiple-choice options from 4 to 10, eliminating trivial questions, and focusing on reasoning-intensive tasks.
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
💡 Pair with ephemeral prompt caching on repository context to cut input spend by >90%.
📊 Claude Sonnet 5.5 Full Benchmark Scorecard
41 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 13 items
BenchCAD variant evaluated with access to a Python tool for programmatic CAD reasoning.
A multilingual benchmark for issue resolving in software engineering that covers Java, TypeScript, JavaScript, Go, Rust, C, and C++.
SWE-Bench Pro is an advanced version of SWE-Bench that evaluates language models on complex, real-world software engineering tasks requiring extended reasoning and multi-step problem solving.
Program Bench evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and documentation.
BenchCAD is a benchmark for programmatic CAD reasoning built from 17,900 execution-verified CadQuery programs spanning 106 industrial part families, roughly half anchored to real ISO, DIN, EN, ASME, and IEC specification tables.
DeepSWE 1.1 evaluates software engineering agents using the mini-swe-agent harness.
Terminal-Bench 4.0 evaluates AI agents on real-world work in terminal and command-line containerized environments across 66 tasks, with emphasis on science-adjacent and frontier engineering problems.
FrontierCode 1.1 Extended subset: agentic coding tasks graded for mergeable patches, reported separately from the Main subset.
FrontierSWE V2 evaluates agents on open-ended technical projects and reports mean@5. Kept separate from FrontierSWE V1 dominance-score results, which are not comparable.
Terminal-Bench-Science 0.1 is a Stanford-led community benchmark of 70 tasks drawn from scientific research workflows across the life, physical, earth, mathematical, and engineering sciences.
CursorBench 4.0 evaluates coding agents on long-horizon coding tasks from real Cursor sessions. Version 4.0 adds longer-horizon tasks; scores are not comparable with CursorBench 3.2.
SWE-Bench Multimodal extends SWE-Bench to evaluate language models on software engineering tasks that involve visual inputs such as screenshots, UI mockups, and diagrams alongside code understanding.
FrontierCode 1.1 evaluates whether coding-agent changes are mergeable, using unit tests, maintainer-defined rubrics, and verifiers. Runs flagged for unfair internet use receive a zero score.
🤖 Agents & Tools 12 items
Partial-credit computer-use score on OSWorld 2.1, as reported in the Claude Sonnet 5.5 release. Separate from OSWorld 2.0 and strict completion scores.
Verified Toolathlon evaluation of agent tool-use capability, as reported in the MiMo-V2.6 release.
OfficeQA is a Databricks public benchmark that evaluates end-to-end grounded reasoning over a large corpus of historical U.S. Treasury Bulletin documents. Models must locate relevant tables across the corpus and perform precise numerical reasoning over them.
LatchBio SpatialBench Verified tests analysis of spatial transcriptomics data across externally validated problems, requiring the model to answer biological questions about the sample from analysis results.
OfficeQA Pro evaluates AI models on professional knowledge-work questions and tasks drawn from real office workflows, including document analysis, spreadsheet reasoning, and information synthesis across business domains.
PhysicianBench is a public benchmark of 100 physician tasks carried out in an electronic health record (EHR), measuring agentic clinical task completion.
Version 2.1 of the Artificial Analysis GDPval evaluation of professional knowledge work, reported as an Elo rating.
AA-Briefcase v1.1 evaluates multi-hour professional office work and reports Elo scores. Kept separate from unversioned AA-Briefcase results.
LatchBio SingleCellBench tests analysis of single-cell RNA sequencing data across problems spanning standard workflows such as labeling cell types, finding differentially expressed genes, and correcting batch effects.
AutomationBench is a tool-use benchmark that evaluates AI agents on automating real-world workflows, testing their ability to orchestrate tools and complete multi-step automation tasks.
Strict pass rate on OSWorld 2.1: fraction of tasks where every graded checkpoint is satisfied. Separate from the OSWorld 2.1 partial-credit score and from OSWorld 2.0.
The Legal Agent Benchmark (LAB) is Harvey's open-source benchmark for evaluating AI agents on complex, long-horizon legal work.
📐 Math 3 items
ArXivMath is a final-answer benchmark of research-level mathematics maintained by MathArena. Problems are extracted monthly from recent arXiv paper abstracts, then filtered through automated and manual checks to ensure they are self-contained, non-trivial, and verifiable.
Full multimodal Humanity's Last Exam evaluated with tool use enabled, distinct from its text-only subset.
Text-only Humanity's Last Exam variant evaluated without tool use.
👁️ Multimodal 4 items
Chartography evaluates professional chart understanding with 100 expert-written tasks drawn from real charts across science, engineering, finance, manufacturing, supply chain, and healthcare.
Biomedical image analysis: quantitative analyses of biomedical images using domain tools.
Chartography visual chart recognition evaluated without tools, distinct from tool-assisted results.
Axiom Bio morphology-to-molecule matching: match multi-channel cell imaging data from a human liver in-vitro assay to shuffled molecule-dose perturbations.
🧠 Reasoning & Science 6 items
A comprehensive multilingual benchmark covering 42 languages that addresses cultural and linguistic biases in evaluation, with improved translation quality and culturally sensitive question subsets.
Multi-task Indic Language Understanding (MILU) covers 11 languages—10 Indic languages (Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, and Telugu) and English—and tests culturally grounded knowledge comprehension.
BioMysteryBench evaluates a model's ability to reason through challenging molecular biology problems, reporting performance on a hard subset and on the subset of problems solved by human experts.
Molecular biology Protocols Troubleshooting variant: detect and fix issues in laboratory protocols.
Medicinal chemistry ADME evaluation: 504 questions asking which real drug-discovery compounds have the highest or lowest measured absorption, distribution, or metabolism-related property.
Protein design Sequence Generation task: generate novel protein sequences conditioned on design constraints.
🏥 Domains 3 items
HealthBench Professional score before length adjustment. Distinct from the length-adjusted HealthBench Professional catalog metric.
HealthBench Professional evaluates model capability and safety for clinician use cases using real clinician-style chats and physician-authored grading rubrics.
An open-source benchmark for measuring performance and safety of large language models in healthcare, consisting of 5,000 multi-turn conversations evaluated by 262 physicians using 48,562 unique rubric criteria across health contexts and behavioral dimensions.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
Closest Self-Hostable Open-Weight Alternatives to Claude Sonnet 5.5
Open-weight models with full HuggingFace weights for private GPU / Mac deployment
Frequently Asked Questions
Is Claude Sonnet 5.5 open-source or proprietary?
Claude Sonnet 5.5 (claude-sonnet-5-5) is a proprietary commercial model by Anthropic. Weights and internal layer dimensions are undisclosed; it is accessed via cloud API.
What are the context limit and API pricing of Claude Sonnet 5.5?
Claude Sonnet 5.5 supports 1000K input context and 128,000 max output tokens, priced at $2/1M input ($0.1/1M cached) and $10/1M output.