claude-3.7-sonnet-20250219

Claude 3.7 Sonnet is Anthropic's hybrid reasoning model, offering an "extended thinking" mode — a configurable chain-of-thought reasoning process that the model applies before producing a final answer. In standard mode it behaves like a fast, capable chat model; in extended thinking mode it applies deeper reasoning for complex tasks. This dual-mode design makes it unusually flexible: a single model can serve both latency-sensitive interactive applications (standard mode) and accuracy-critical reasoning tasks (extended thinking), without requiring separate model deployments.

Claude 3.7 Sonnet is priced at $3.00 per million input tokens and $15.00 per million output tokens in standard mode. Extended thinking mode incurs additional cost for the thinking tokens generated internally — these are billed at the output token rate and can add $5–$50+ per query for complex tasks depending on thinking budget. Prompt caching reduces repeated input to $0.30/1M. At 1,000 input + 500 output tokens per call in standard mode, it costs approximately $0.0105 per 1,000 API calls — the same as Claude 3.5 Sonnet. The extended thinking cost makes it significantly more expensive for reasoning-heavy workloads, but the quality improvement on hard tasks often justifies the premium.

Claude 3.7 Sonnet supports a 200K-token context window in both standard and extended thinking modes. Standard mode latency is approximately 400–600ms TTFT, similar to Claude 3.5 Sonnet. Extended thinking mode latency varies from 5 to 120+ seconds depending on the thinking budget — the `thinking.budget_tokens` parameter controls how many tokens the model can use for internal reasoning, with higher budgets enabling deeper analysis at the cost of longer wait times. Maximum output is 64,000 tokens in extended thinking mode (vs 8,192 in standard mode), enabling very long-form generation for complex tasks. Throughput in standard mode is approximately 80 tokens per second.

Claude 3.7 Sonnet is best suited to coding tasks requiring multi-step reasoning (it achieves top scores on agentic coding benchmarks in extended thinking mode), complex mathematical analysis, scientific reasoning, and any task where standard models produce shallow or incorrect answers. In standard mode it is a strong general-purpose model suitable for the same workloads as Claude 3.5 Sonnet. The extended thinking mode is particularly valuable for: debugging complex code across large codebases, solving multi-step mathematical problems, analyzing ambiguous requirements and producing structured plans, and any task where the model needs to "work through" a problem rather than pattern-match to a known answer. Computer use capability (beta) is also available.

Claude 3.7 Sonnet sits at the top of Anthropic's lineup alongside Claude Opus 4, with the key distinction being that Opus 4 is Anthropic's most capable model overall while 3.7 Sonnet offers the best cost-to-capability ratio for reasoning tasks. Its main competitors are extended-reasoning models from OpenAI and Google at similar price points. For teams already using Claude 3.5 Sonnet, upgrading to 3.7 Sonnet is straightforward — the API interface is identical, and the extended thinking mode is opt-in via an additional parameter. The intelligence score of approximately 90/100 reflects near-frontier capability, particularly on coding and reasoning benchmarks.