Claude Sonnet 5.5
Pricing & Specs
Anthropic · $2 input / $10 output per 1M tokens · 1M context window
The per-token price never tells the full story. A typical task (1K input + 500 output tokens) on Claude Sonnet 5.5 costs about $0.0070, roughly $7.00 per 1,000 runs. But if it needs 3x the tokens of a cheaper model to match quality on YOUR task, the economics flip. The only way to know is to benchmark it on your actual workload.
Claude Sonnet 5.5 API Pricing
| Tokens | Price |
|---|---|
| Input | $2 / 1M tokens |
| Output | $10 / 1M tokens |
| Cached input | $0.2 / 1M tokens |
50% discount available for batch API requests. Cache writes: $2.50/MTok (5min), $4/MTok (1hr). Cache reads: $0.20/MTok. Min cacheable prompt 512 tokens.
What that means in practice
Claude Sonnet 5.5 Specs
| Spec | Value |
|---|---|
| Context window | 1M tokens |
| Max output tokens | 128K tokens |
| Input modalities | text, image |
| Output modalities | text |
| Knowledge cutoff | Jun 2026 |
| Reasoning model | Yes |
| Tool / function calling | Yes |
| JSON mode | Yes |
| Streaming | Yes |
| Prompt caching | Yes |
| Batch API | Yes |
Notes
Released September 28, 2026. Adaptive thinking on by default, default effort high; lowest setting is between_tools (turns off up-front thinking, high effort or below). Non-default temperature/top_p/top_k returns 400. Breaking vs Sonnet 5: forced tool use returns error; thinking blocks tied to producing model; text between tool calls returned in thinking blocks.
FAQ
How much does Claude Sonnet 5.5 cost?
Claude Sonnet 5.5 costs $2 per 1M input tokens and $10 per 1M output tokens ($0.2 per 1M cached input tokens).
What is Claude Sonnet 5.5's context window?
Claude Sonnet 5.5 has a 1M-token context window. Maximum output is 128K tokens.
Can I test Claude Sonnet 5.5 on my own task?
Yes. OpenMark lets you benchmark Claude Sonnet 5.5 against 100+ models on your own task with real API calls, with no API keys needed, free tier available.
Is Claude Sonnet 5.5 the right model for YOUR task?
Pricing tables can't answer that. Benchmark it against 100+ models
on your actual workload: real API calls, real costs. Free tier available.