Claude Sonnet 4.6
Claude Sonnet 4.6 is a text-and-image model from Anthropic in the Claude Sonnet line, offering a 1,000,000-token context window and an August 2025 knowledge cutoff.
API Pricing
Cheapest on Velokey — 48% below avg| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $1.20 | $6.00 | $0.120 | |
| $1.50 | $7.50 | $0.150 | |
| $1.50 | $7.50 | $0.150 | |
| $3.00 | $15.00 | - | |
| $3.00 | $15.00 | $0.300 | |
| $3.00 | $15.00 | $0.300 | |
| $3.00 | $15.00 | $0.300 |
Prices updated daily. Last check: Sep 25, 2026
Claude Sonnet 4.6 pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Reasoning & Knowledge
- GPQA Diamond79.9%
- Humanity's Last Exam13.3%
Agentic & Tool Use
- Terminal-Bench Hard46.2%
- τ²-bench79.5%
Instruction & Long Context
- IFBench41.2%
- Long-Context Reasoning68.3%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Anthropic
- Family
- Sonnet
- Tier
- Flagship
- Context Window
- 1.0M
- Knowledge Cutoff
- Aug 2025
- Modalities
- Text, Image
Capabilities
- Tool Calling
- No
- Open Source
- No
- Aliases
- claude-sonnet-4-6, claude-sonnet-4-6-20260217, Claude Sonnet 4.6
Strengths & Limitations
Strengths
- 1,000,000-token context window, large enough to hold entire codebases or long document collections in one request
- Accepts image input alongside text, enabling screenshot, diagram, and scanned-document workflows
- Knowledge cutoff of August 2025, more recent than many widely deployed models
- Sonnet-tier positioning targets a middle ground between Anthropic's Haiku and Opus lines for production workloads
- Dated snapshot alias (claude-sonnet-4-6-20260217) allows pinning to a specific version for reproducible deployments
- Measured time to first token around 1.45 seconds in third-party testing, suitable for interactive chat and IDE integrations
- Available through multiple hosting providers, so pricing and throughput can be compared side by side
Limitations
- Measured output throughput of roughly 47 tokens per second is moderate; latency-critical, high-volume tasks may favor a smaller Haiku-tier model
- Filling the 1M-token context window is token-intensive, and long-context requests can be substantially more expensive per call than short prompts
- Model weights are not distributed for self-hosting — access is through hosted APIs
- Knowledge cutoff of August 2025 means events after that date require retrieval, search, or explicit context
- Benchmark scores for reasoning, coding, and math are not tracked in our database for this model, so capability comparisons must be drawn from other sources
Key Features
About Claude Sonnet 4.6
Common Use Cases
Claude Sonnet 4.6 fits workloads that need substantial reasoning capability at a mid-tier cost point: software engineering assistants that must read across many files at once, contract and report analysis over hundreds of pages, and agentic pipelines where a long trace of tool calls and intermediate results accumulates in context. The 1M-token window makes it a reasonable choice for tasks where building a retrieval layer is more trouble than simply passing the full corpus, such as one-off codebase audits, migration planning, or reviewing a large set of related documents for consistency. Image input extends this to mixed-media inputs — UI screenshots, architecture diagrams, charts, and scanned forms. For very high-volume classification, routing, or extraction where per-token cost and latency dominate, a smaller Haiku-tier model is often the better economic fit; for the hardest multi-step reasoning problems, Anthropic's Opus tier is the usual step up.
Frequently Asked Questions
How much does Claude Sonnet 4.6 cost?
Pricing depends on the provider you use and the pricing type — input tokens, output tokens, cached reads, and batch processing are typically billed at different rates, and long-context requests can carry different pricing than short ones. Because rates change frequently and vary between hosts, check the pricing table on this page for current figures rather than relying on a fixed number.
What is Claude Sonnet 4.6 best used for?
It suits coding assistance, long-document and repository analysis, and agentic workflows that accumulate large amounts of context. The 1,000,000-token window lets you pass full codebases or document sets directly, and image input covers screenshots, diagrams, and scanned pages alongside text.
How does it compare to Anthropic's Haiku and Opus tiers?
Anthropic's lineup places Sonnet between Haiku and Opus. Haiku models target high-volume, latency-sensitive tasks at lower cost; Opus models target the most demanding reasoning work. Sonnet 4.6 is positioned as the general-purpose production option — capable enough for complex coding and analysis while cheaper than Opus. Compare live rates for each in the table above.
Can Claude Sonnet 4.6 process images?
Yes. The model accepts both text and image input, so screenshots, charts, diagrams, and scanned documents can be included in a request alongside text.
How current is the model's knowledge?
Its training knowledge extends to roughly August 2025. For anything after that date, supply the information in the prompt or connect the model to a search or retrieval tool.
How fast is it?
Third-party measurements from Artificial Analysis report around 46.8 output tokens per second and a time to first token of about 1,446 ms. Actual figures vary by provider, prompt length, and load, so treat these as approximate.