Claude 3.7 Sonnet
Era 5 · Generative AI, reasoning, and agents (2022–2026) › Alignment and reasoning
- Lineage
- Symbols and logic
- Kind
- Implementation
Who
Anthropic
What happened
Anthropic releases Claude 3.7 Sonnet, which it presents as the first hybrid reasoning model on the market: it answers almost instantly or thinks step by step, as the user chooses. Alongside it, Anthropic launches a preview of Claude Code, its first agentic coding tool.
Why it mattered
Extended reasoning becomes an option within the same model, with a thinking budget that can be adjusted.
The full story
After o1 and DeepSeek-R1, reasoning models were separate models: good at mathematics and coding, but slower and more expensive for simple tasks. Anyone who wanted both had to choose between different models.
Claude 3.7 Sonnet brought them together in one. In standard mode it worked as an improved version of Claude 3.5 Sonnet. In extended thinking mode it reflected before answering, which improved its results in mathematics, physics, instruction following, and coding. API users could set how many tokens, the chunks of text the model works with, it could spend on thinking, up to a maximum of 128,000, and so balance speed, cost, and quality.
On SWE-bench Verified, a benchmark of real programming issues taken from code repositories, it scored 62.3 percent, and 70.3 percent with a custom scaffold, the program that organizes the model's work on the code. It kept the price of its predecessors.
The same day Anthropic presented Claude Code, its agentic coding tool, whose story is told in the agents field.
Builds on
OpenAI o1 (September 12, 2024)
Inherited, with no proven direct link
After o1, the three labs add to their main models a mode that reasons before answering
Threads that pass through here
- From prompted reasoning to trained reasoning (step 8 of 9)
Sources
Original source
How to cite
“Claude 3.7 Sonnet.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/claude-3-7-sonnet
This entry was last reviewed on November 13, 2026.