Curiosity · AI Model
Claude 3.7 Sonnet
Claude 3.7 Sonnet is the February 2025 model where Anthropic introduced extended thinking — a switchable mode that lets the same model respond instantly for easy prompts and deliberate for up to 128K reasoning tokens on hard ones. It was Anthropic's first hybrid-reasoning model and the bridge between the Claude 3 and Claude 4 families.
Model specs
- Vendor
- Anthropic
- Family
- Claude 3.7
- Released
- 2025-02
- Context window
- 200,000 tokens
- Modalities
- text, vision, code
- Input price
- $3/M tok
- Output price
- $15/M tok
- Pricing as of
- 2026-04-20
Strengths
- Introduced extended thinking — a single model for chat and reasoning
- Launched Claude Code CLI, setting the template for agent coding
- Strong coding scores at launch, briefly SOTA on SWE-bench Verified
- Tunable reasoning budget via API
Limitations
- Context window capped at 200K — Claude 4 models now reach 1M
- Extended-thinking tokens billed as output, inflating bills
- Superseded by Claude 4 family across nearly every benchmark
Use cases
- Coding agents — powered Claude Code at launch
- Problem-solving apps that benefit from toggleable reasoning
- Research and analysis where depth > latency on some prompts
- Agent pipelines that need both fast and deliberative steps
Benchmarks
| Benchmark | Score | As of |
|---|---|---|
| SWE-bench Verified | ≈62% | 2025-02 |
| AIME 2024 (math, thinking) | ≈80% | 2025-02 |
| GPQA Diamond | ≈68% | 2025-02 |
Frequently asked questions
What is Claude 3.7 Sonnet?
Claude 3.7 Sonnet is Anthropic's February 2025 model and the first Claude with extended thinking — a switchable reasoning mode that mixes fast responses and long chain-of-thought in one set of weights.
What is extended thinking in Claude 3.7 Sonnet?
Extended thinking is an API parameter that lets 3.7 Sonnet produce a long internal reasoning trace (up to 128K tokens) before its final answer. For easy prompts the model answers immediately; for hard ones it can think much longer, improving accuracy.
Is Claude 3.7 Sonnet still worth using?
For new projects, Claude Sonnet 4.6 or Sonnet 4.5 is preferable. Claude 3.7 Sonnet is still available on the API for teams that depend on its exact behaviour, but the Claude 4 family improves on it across nearly all benchmarks.
How much does Claude 3.7 Sonnet cost?
As of April 2026, Claude 3.7 Sonnet is priced at USD 3 per million input tokens and USD 15 per million output tokens, the same headline pricing as Sonnet 4.5 and 4.6.
Sources
- Anthropic — Claude 3.7 Sonnet and Claude Code — accessed 2026-04-20
- Anthropic — Pricing — accessed 2026-04-20