Curiosity · AI Model
Claude Sonnet 4.6
Claude Sonnet 4.6 is the production workhorse in Anthropic's 2026 Claude 4 family. It sits below Opus on cost but keeps nearly all of Opus's coding and tool-use reliability, with a 1M-token context and extended thinking. For most developers using Claude in production, Sonnet 4.6 is the default model, with Opus reserved for the hardest tasks.
Model specs
- Vendor
- Anthropic
- Family
- Claude 4
- Released
- 2026-01
- Context window
- 1,000,000 tokens
- Modalities
- text, vision, code
- Input price
- $3/M tok
- Output price
- $15/M tok
- Pricing as of
- 2026-04-20
Strengths
- Best cost/quality ratio in Anthropic's lineup
- Very reliable tool use and structured outputs
- 1M context plus prompt caching cuts long-document costs sharply
- Follows XML-tag prompts precisely — great for templated pipelines
Limitations
- Trails Opus 4.7 on frontier reasoning and hardest coding tasks
- Slower than Haiku 4.5 for ultra-low-latency UX
- No audio modality — pair with OpenAI Realtime or Gemini Live for voice
Use cases
- Production coding agents — the default Claude Code model
- Customer support and product assistants at scale
- Multi-step agents with MCP servers and tool use
- Content generation and summarisation for long documents
Benchmarks
| Benchmark | Score | As of |
|---|---|---|
| SWE-bench Verified | ≈70% | 2026-04 |
| MMLU-Pro | ≈84% | 2026-04 |
| GPQA Diamond | ≈66% | 2026-04 |
Frequently asked questions
What is Claude Sonnet 4.6?
Claude Sonnet 4.6 is the mid-tier model in Anthropic's 2026 Claude 4 family. It balances cost, latency, and quality — typically used as the default production model while Opus 4.7 is reserved for the hardest workloads.
When should I pick Sonnet 4.6 over Opus 4.7?
Choose Sonnet 4.6 for nearly all production workloads — coding copilots, support agents, RAG, tool-calling. Pick Opus 4.7 only when quality noticeably improves your end-user result and cost is tolerable.
How much does Claude Sonnet 4.6 cost?
As of April 2026, Sonnet 4.6 is priced at roughly USD 3 per million input tokens and USD 15 per million output tokens on the Anthropic API. Prompt caching further reduces effective cost.
Does Sonnet 4.6 support extended thinking?
Yes. Sonnet 4.6 supports Anthropic's extended-thinking mode, where the model produces a long internal reasoning trace before its final answer. This is toggled via an API parameter.
Sources
- Anthropic — Models overview — accessed 2026-04-20
- Anthropic — Pricing — accessed 2026-04-20